Received: 25 June 2024 Accepted: 25 January 2025

DOI: 10.1002/agj2.70028

ORIGINAL ARTICLE
S o i l Fe r t i l i t y a n d C ro p Nu t r i t i o n

Soil test phosphorus predicts field-level but not subfield-level corn

yield response

Vaughn Reed1 Jenni Fridgen2 Bronc Finch3 John Spargo4

Josh McGrath5 James M. Bowen6 Gene Hahn7 Douglas Smith8 Edwin Ritchey9
1 Department of Plant and Soil Sciences, Mississippi State University, Mississippi State, Mississippi, USA
2 Nutrien Ag Solutions, Nicholasville, Kentucky, USA
3 Soil Fertility, Crop, Soil, and Environmental Sciences, University of Arkansas, Little Rock, Arkansas, USA
4 Agricultural Analytical Services Lab, Department of Plant Science, Penn State University, University Park, Pennsylvania, USA
5 Office of National Programs, USDA ARS, Beltsville, Maryland, USA
6 Department of Plant and Soil Sciences, University of Kentucky, Lexington, Kentucky, USA
7 Department of Plant Science & Landscape Architecture, University of Maryland, College Park, Maryland, USA
8 Grassland Soil and Water Research Laboratory, Temple, Texas, USA
9 Department of Plant and Soil Sciences, University of Kentucky, Princeton, Kentucky, USA

Correspondence
Vaughn Reed, Department of Plant and Soil Abstract
Sciences, Mississippi State University, Soil test-based fertilizer recommendations traditionally serve to predict average nutri-
Mississippi State, MS 39762, USA.
Email: vr401@msstate.edu ent needs across fields, but their effectiveness for precision agriculture remains
uncertain. Our objectives were to evaluate whether soil phosphorus (P) concentra-
Assigned to Associate Editor Davie
tions predicted corn (Zea mays,r L.) yield response to P at the sub-field level, and
Kadyampakeni.
to determine if soil test critical levels varied within field boundaries. We conducted
research over seven growing seasons at two Kentucky sites collecting spatially dense
yield response data from over 150 paired plots per field. Mehlich 3 extractable phos-
phorus (M3P) soil ranged from 0.8 to 63 mg kg−1 , with 96% of sample points falling
below the University of Kentucky’s fertilizer cutoff of 30 mg kg−1 M3P for corn.
Each plot (10−2 ha) received 0 or 29.5 kg ha−1 P. While M3P effectively predicted
average field-level response, with yield increases in five of seven site-years, it failed to
predict subfield responses, where only 51% of plots showed positive yield response to
P application. Linear plateau models revealed that conventional statistical treatments
of soil test correlation data mask important subfield variability. The poor relationship
between soil test P and yield response at the subfield scale suggests that variable rate
P management requires incorporating additional factors beyond soil P concentration
or moving away from such deterministic models toward probabilistic models. Our
findings demonstrate that while current soil test recommendations provide accurate
field-scale guidance, they lack the precision required for variable rate application.

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original
work is properly cited.
© 2025 The Author(s). Agronomy Journal published by Wiley Periodicals LLC on behalf of American Society of Agronomy.

Agronomy Journal. 2025;117:e270028. wileyonlinelibrary.com/journal/agj2 1 of 12

https://doi.org/10.1002/agj2.70028
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 of 12 REED ET AL.

1 INTRODUCTION
Core Ideas
Current fertilizer prices and environmental concerns dictate
∙ Soil test phosphorus effectively predicts field-level
that we increase fertilizer use efficiency. Precision nutrient
management seeks to increase nutrient efficiency by match- but not subfield-level corn yield response.
∙ Variable rate phosphorus management requires
ing nutrient inputs to spatially and temporally variable crop
nutrient needs. Individual plants rely on soil nutrient supply, models beyond traditional soil test correlation.
∙ High-density, paired response plots revealed spa-
as a function of amount and intensity, and external fertilizer
inputs to meet their internal nutrient requirement. Soil testing tial variability in phosphorus response not captured
provides the basis for conventional phosphorus (P) fertilizer by current methods.
management. Soil test correlation estimates soil nutrient
supply by relating extractable soil P concentration to relative
crop response to P application. These correlation procedures
establish critical soil test concentrations or ranges, above This study proposes a novel soil test correlation field study,
which we do not expect a crop response to added phosphorus. designed to support variable rate fertilizer applications. Our
Soil test calibration predicts the fertilizer rate needed to study aimed to evaluate the effectiveness of soil test P in
deliver the balance of the plant’s nutrient need and reach the predicting crop nutrient response at sub-field levels and to
maximum obtainable yield at a given soil P concentration examine spatial variability in soil test critical levels within
(Pearce et al., 2022). fields. We hypothesized that existing soil test P critical lev-
Recent studies have shown that current soil test nutrient els for Kentucky, designed for broad-scale accuracy, would
recommendations are generally accurate; however, recom- not reliably predict P response within field boundaries due to
mendations need more work to improve their precision by a lack of site-specific precision.
reflecting modern practices, higher yields, improved crop
genetics, and spatially variable growing environments (Hop-
kins & Hansen, 2019; Reed et al., 2021). Often, current 2 MATERIALS AND METHODS
fertilizer recommendation systems in the United States rely
on correlation and calibration data generated in the mid- 2.1 Experimental design, crop
20th century (Lyons et al., 2021). These studies provided management, and site description
average critical soil test concentrations or ranges at state or
regional scales (median state area of 14.4 × 106 ha) (USDA- We established this study in 2016 at two sites, one located near
NASS, 2024) using soil test extractants appropriate for their Princeton, KY (37.112, −87.267) in Caldwell County, and
regions, such as the Mehlich 3, Lancaster, Olsen, or Bray tests, the other near Quicksand, KY (37.535, −83.346) in Breathitt
(Dari et al., 2019; Sikora & Moore, 2014) to support con- County (Figure 1). Two grass waterways divided the Prince-
ventional flat-rate nutrient recommendations. Our ability to ton field into three sections (Figure 2), which totaled 4.854 ha,
develop more spatially precise recommendations from his- and contained a Zanesville silt loam (fine-silty, mixed, active,
toric trial data are limited because scientists often did not and mesic Oxyaquic Fragiudalfs). The Quicksand field con-
include complete plot-level data in their publications (Slaton sisted of 2.55 ha (Figure 3) and contained a Chagrin–Grigsby
et al., 2022). Complex (fine-loamy mixed, active, mesic Dystric Fluventic
The spatial variability of nutrient supply complicates gen- Eutrudepts and coarse-loamy, mixed, active, and mesic Dys-
eration of precise nutrient recommendations. Studies have tric Fluventic Eutrudepts). Prior to study initiation, University
shown that nutrient concentrations can vary across space both of Kentucky (UKY) managed both sites in no-till systems,
vertically (Hansel et al., 2017; Howard et al., 1999; Souza, producing continuous corn (Zea mays, L.) at Quicksand, and
2020) and horizontally (Solie et al., 1999). In addition, soil a corn, winter wheat (Triticum aestivum, L.), and soybean
properties, chemical, physical, and biological, that influence [Glycine max, (L.), Merr.] rotation at Princeton for >10 years.
nutrient availability, such as soil texture, density, pH and Precise fertilizer records were not available for either site
organic matter, also vary at the field scale (Mzuku et al., prior to 2016. In general, the Quicksand site had a history of
2005). Although clear data on the prevalence of variable rate only urea fertilizer application without liming, and the Prince-
fertilizer application do not exist, it has clearly expanded in ton site had routine N fertilizer application, and sporadic K,
conventional grain production (Pierpaoli et al., 2013). A need P, and lime applications. Prior to this study, UKY managed
exists to develop nutrient recommendations that precisely and these fields for commercial grain production, not research,
accurately meet spatially variable crop nutrient needs. and applied nutrients uniformly across the fields.
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
REED ET AL. 3 of 12

F I G U R E 1 Map Depicting the two sites from this study. The “PR” label and point depicts the Princeton site, and the “QS” label and point
depicts the Quicksand site. Inset shows the site of Kentucky inside the contiguous United States.

F I G U R E 2 Map Depicting main plots across the Princeton site in Princeton, KY. Plots established in 2016 were 9 × 9 m, and data were
collected in 2016, 2018, and 2020. Plots established in 2018 were 12.2 × 12.2 m, and data were collected in 2018 and 2020.

During the study (2016–2021), both fields followed a corn- plot into three subplots, measuring 3 m by 9 m (Figure 4A).
soybean crop rotation, except for wheat-double crop soybeans We planted corn and soybean in rows 76 cm apart, and wheat
at Princeton-2017 and corn after corn at Quicksand 2020– was drilled in rows 19 cm apart. We assigned the control treat-
2021 (Table 1). Researchers applied lime to the Quicksand ment (0 kg ha−1 P) to each edge subplot, and the P treatment
site in 2019 prior to soybean planting to address declining (29.5 kg ha−1 P) to the center subplot. In 2018, we added
soil pH. This manuscript only contains data collected from 54 and 55 main plots to the Quicksand and Princeton sites,
the corn portions of the crop rotation. respectively. Due to space limitations, the main plots added
Before trial establishment in 2016, we overlaid each field to Quicksand followed the same three-subplot scheme as the
with a 9-m grid using GIS software, and randomly selected 2016 season. However, because Princeton had more free space
123 and 101 grid cells as main plots at Quicksand and Prince- available, the new main plots in a 12.2 m grid included four
ton, respectively (Figures 2 and 3). We then divided each main subplots (Figure 4B), which allowed randomization of the
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4 of 12 REED ET AL.

FIGURE 3 Map Depicting main plots across the Quicksand site in Quicksand, KY. Each main plot was 9 × 9 m.

T A B L E 1 Cropping system employed at each site across all years. application. Mechanical application dictated that we use liq-
For Princeton 2017, the winter wheat crop was planted after corn uid fertilizer to allow precise rate changes over the shortest
harvest in 2016, and harvested in June 2017, followed by establishment distance possible. Through expert consensus, we decided that
of soybean crop. subsurface band application offered the highest probability of
Year Quicksand (crop) Princeton (crop) seeing a corn yield response to P under no-till management.
2016 Corn Corn We planted corn and applied the starter P and N treatments
2017 Soybean Winter wheat/double crop with a four-row (76.2 cm row) not-till planter (John Deere
soybean MaxEmerge Plus 1750) outfitted with two electric variable
2018 Corn Corn rate pumps (SureFire Ag Systems Tower Fertilizer System for
2019 Soybean Soybean Field-IQ-PWM Control), one for APP and one for UAN. Tee-
2020 Corn Corn
Jet solenoid nozzles (TeeJet, 12 V e-ChemSaver) applied the
fertilizer behind Yetter fertilizer coulters. We achieved pre-
2021 Corn Soybean
cise rate control using Geographic Position System (GPS)
with Real Time Kinetic (RTK) correction (Trimble, FmX Fm
1000 internal receiver), a Trimble FmX display, and a Trimble
control and P treatment with two replicates of each within the FieldIQ controller.
main plot. We dribbled 396 L ha−1 UAN in between corn rows at V6
During corn planting, we applied liquid fertilizer in a band growth stage to provide an additional 168 kg ha−1 N, for a total
approximately 5 cm beside and 5 cm below the seed (collo- season rate of 224 kg ha−1 N. The UKY guidelines (Ritchey
quially referred to as 2 × 2). The control treatment received & McGrath, 2021) recommend 15-98 kg ha−1 P for soils test-
urea-ammonium nitrate (UAN, 32-0-0) at a rate of 133 L ing less than or equal to the critical concentration of 30 mg
ha−1 to provide 56 kg ha−1 N. The P treatment received 86 kg−1 M3P (where M3P is Mehlich 3 extractable phosphorus).
L ha−1 of UAN and 142 L ha−1 of ammonium polyphosphate Both Princeton and Quicksand had low average M3P concen-
(APP, 10-34-0) to provide 56 kg ha−1 N and 29.5 kg ha−1 trations that would receive a recommendation of up to 39 kg
P. A group of soil fertility experts from industry, academics, ha−1 P for surface broadcast P fertilizer. Therefore, we antic-
and government discussed the best way to apply fertilizer for ipated that banded 29.5 kg ha−1 P would provide more than
this trial. These conversations occurred informally and for- adequate P to generate a positive yield response where M3P
mally as part of the Mule Barn meetings (Osmond et al., was <30 mg kg−1 (Ritchey & McGrath, 2021). This exper-
2024). The size of the fields and number of plots required imental design sought to map P response, rather than attain
automated, mechanical fertilizer application instead of hand maximum yield response to P or to build soil P concentrations.
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
REED ET AL. 5 of 12

F I G U R E 4 The plot design incorporated two subplots at both sites in 2016 (A), with the center plot (B) receiving phosphorus fertilizer and the
two edge plots (A and C) receiving no phosphorus. In 2018, plots added at Quicksand used the original design (A). Plots added at the Princeton site
used a four-subplot design (B) with two subplots randomly selected as controls and the other two receiving phosphorus fertilizer.

Based on our experiences, the banded P rate would generate a We harvested the center two rows (total area of 13.5 m2 for
significant P response on average across soils within the range 9 × 9 m main plots, 18.3 m2 for 12.2 × 12.2 m main plots)
of concentrations seen at both sites. of every four-row planter pass using a Kincaid 8-XP plot
combine (Kincaid Equipment Manufacturing) with a two-row
corn head. Yield was estimated from impulse and moisture
2.2 Soil and crop data measurements taken by an AgLeader sensor plate and mois-
ture model (AgLeader Technology), which were then logged
We used a Wintex 1000 automated soil probe (Wintex Agro) to an AgLeader Insight display, along with position and speed,
to collect soil samples (0- to 10-cm depth, 18-mm diameter) determined by RTK-corrected GPS. Grain yield was adjusted
from each subplot prior to crop establishment in the 2016, to reflect industry standard 15.5% moisture content.
2018, and 2021 growing seasons at both sites. Each sam- We calculated RY for each main plot by dividing the con-
ple consisted of 10 individual soil cores collected randomly trol treatment (average of the subplots without P fertilizer)
throughout the subplot and thoroughly mixed to form a com- by the P treatment (or average of P treatment subplots where
posite sample. Soil samples were dried at 38˚C for 24 h, replicated). Publications reporting linear plateau datasets may
ground to pass a 2-mm sieve, and submitted to the UKY sometimes choose to cap relative yield values at 1.0 (Pearce
Soil Testing Laboratory (UKSTL). The UKSTL determined et al., 2022). Relative yield can exceed 1.0 in plots where
soil pH (1 M KCl with glass electrode), Sikora Buffer pH, the control plots will yield greater than the plots with fertil-
M3P, and Mehlich 3 extractable potassium using Inductively izer application. In this study, we investigated the correlation
Coupled Plasma (ICP) methods from established soil testing using both constrained and unconstrained relative yield val-
protocols (Sikora & Moore, 2014). Although we sampled each ues. The discrepancy with these values can directly impact
subplot individually, this manuscript correlates the average of the models used to depict yield response. We conducted soil
the soil test results from the control plots against relativized test correlation by regressing the constrained or unconstrained
yield (RY). Therefore, we do not report changes in soil test relative yields for each main plot against the mean M3P con-
P over time in the fertilized subplots. To calculate average centration of the unfertilized subplots in that main plot to
pH for the control subplots, we first transformed subplot pH determine the critical soil test concentration above which no
values to hydrogen concentration [H+ ], averaged the concen- further yield response was likely. Since we did not collect soil
trations, and then transformed that value to pH by taking the samples in 2020, we used results from 2021 soil sampling for
inverse logarithm of the [H+]. the correlation procedures.
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6 of 12 REED ET AL.

FIGURE 5 Density plots of Mehlich 3 extractable phosphorus (P) and soil pH of main plots (average of control treatment subplots) across both
sites and years.

Our objective was to understand the spatial variability in determine if M3P was a predictor of nutrient response, we
nutrient response across site years. Therefore, the original conducted a generalized linear model for binomial response
plot design allowed estimation of the yield response to P fer- (yes or no) and regressed against M3P concentration similar
tilizer at the main plot scale. As described previously, the to the previous procedure.
original design did not replicate the P fertilized treatment We performed non-linear modeling with SAS software,
within main plots, and therefore, did not allow error estima- Version 9.4, using PROC NLIN (SAS Institute). Non-linear
tion for response at the main plot level. We determined that analysis was conducted across both sites and by site, with
a main plot was responsive to P if the relative yield value constrained and unconstrained relative yield values. Dif-
was <0.95. ferences between non-linear models were determined by
computing the F-statistic from the difference of the sum
of squares of the site specific and all sites, divided by the
2.3 Statistical procedures mean square values of all sites, with an alpha value of
0.10.
All regression and analysis of variance (ANOVA) were per-
formed using R software, Version 4.2.1 (R Core Team, 2022).
Datapoints from grain yield were determined to be outliers 3 RESULTS
and removed if they were >50% outside the interquartile range
(IQR) or 1.5 × IQR of the population from that site-year. We 3.1 Soil phosphorus and pH
conducted an ANOVA (α = 0.10) for grain yield across all
site-years using grain yield as the dependent variable, and For this study, we selected field sites with average M3P con-
M3P as the independent variable, and included site-year and centrations below 30 mg kg−1 (Figure 5; Table 2), the M3P
main plot as random variables to determine if M3P was a pre- concentration below which UKY recommends P fertilizer for
dictor of yield. If there was a significant interaction, we sliced corn (Ritchey & McGrath, 2021). Across all years, main plot
by site-year to determine site-year-specific relationships. To M3P averaged 14 and 12 mg kg−1 and ranged from 1 to
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
REED ET AL. 7 of 12

T A B L E 2 Summary Statistics for Mehlich 3 extractable 3.2 Average phosphorus response

phosphorus (M3P) and soil pH.

Standard Median absolute Phosphorus fertilizer significantly impacted mean yield

Mean error Median deviation across all observations; however, there was a significant inter-
Princeton action by both site and year (data not shown). Therefore,
2016 we sliced the data by both site and year to determine site-
M3P (mg kg−1 ) 19 0.7 19 7.4 and year-specific trends. Phosphorus fertilizer significantly
Soil pH 5.9 0.02 6.2 0.42
(p < 0.0001) increased mean corn yields in 2016 at Princeton
and all years at Quicksand (Table 3). Corn yield at Princeton
2018
was not consistently responsive to P fertilizer, with the only
M3P (mg kg−1 ) 14 0.4 12 4.5
significant (p = 0.0789) response coming from 2016, where
Soil pH 5.5 0.02 5.8 0.42
P fertilizer increased yield 0.46 ± 0.09 Mg ha−1 . On aver-
2021 age, Princeton yielded 7.11 ± 0.09 and 3.70 ± 0.07 Mg ha−1
M3P (mg kg−1 ) 9 0.3 8 3.7 regardless of treatment (data not shown) in 2018 and 2020,
Soil pH 5.3 0.02 5.5 0.44 respectively. Low yields in 2020 were due to poor crop estab-
Quicksand lishment, and hot and dry summer, and the crop was never
2016 able to fully recover. On average, P fertilizer increased Quick-
M3P (mg kg−1 ) 13 0.4 11 4.5 sand corn yield by 0.52 ± 0.10, 0.84 ± 0.11, 0.73 ± 0.224,
Soil pH 4.6 0.04 4.6 0.39 and 0.82 ± 0.16 Mg ha−1 in 2016, 2018, 2020, and 2021,
2018 respectively. Across all years, this represents an average yield
M3P (mg kg−1 ) 12 0.3 10 3.7 increase of 6%.
Soil pH 4.5 0.04 4.60 0.40
2021
3.3 Modeling phosphorus response
M3P (mg kg−1 ) 12 0.3 10 4.45
Soil pH 5.2 0.03 5.40 0.56 Mehlich 3 P was a significant predictor of mean yield
response across all data (p < 0.0001, Table 5). However, with
yield there was a significant interaction with site and year
(p = 0.0077); therefore, the data were sliced by site-year.
At Princeton, yield responded to P fertilizer in 44%, 31%,
63 mg kg−1 and 3 to 47 mg kg−1 at Princeton and Quicksand, and 50% of main plots in 2016, 2018, and 2020, respectively
respectively. Approximately 94% and 98% of plots at Prince- (Table 5). The M3P was below 30 mg kg−1 in 93% of the
ton and Quicksand, respectively, had M3P below the 30 mg responsive plots yet was only a significant predictor of yield
kg−1 threshold. Moreover, the M3P data had a strong positive response in 2020 (p = 0.0045). At Quicksand, yield responded
skew, with most samples falling below 14 mg kg−1 , the upper to P fertilizer in 51%, 53%, 70%, and 56% of main plots in
limit of UKY’s “Low” interpretative category. Across all site 2016, 2018, 2020, and 2021, respectively. The concentration
years, 96% of the plots had M3P below 30 mg kg−1 , where of M3P was <30 mg kg−1 in 99% of the responsive plots yet
UKY recommends P fertilizer application. was only a significant predictor of nutrient responsiveness in
Soil pH impacts soil P availability and potential crop 2016 (p = 0.0038).
response (Penn & Camberato, 2019). The Quicksand site Linear plateau models, where significant, were fitted by
had an average soil pH of 4.6, ranging from 4.0 to 7.3 in site and across sites, with constrained and unconstrained rel-
2016 and 2018. This site was limed in the spring of 2019, ative yield values (Table 4; Figure 6). Data were not robust
which increased the average soil pH to 5.2 (range 4.6–7.2) in enough to provide significant models by year, so all years were
2021. The pH at Quicksand had a positive skew and a small, compiled to provide site-specific models. At Princeton, the
consistent cluster of plots with pH around 7.0. The high pH linear plateau model was not significant when relative yield
cluster was adjacent to a road with gravel shoulders and a was unconstrained (p = 0.3770) but was significant when
limestone (CaCO3 ) gravel aggregate surface at one time. We relative yield was constrained at 1.0 (p = 0.0005) with a
suspect this caused the cluster of high pH plots in that area. joint point of 16.5 mg kg−1 and a pseudo-r2 value of 0.02.
The pH at Princeton was less skewed than at Quicksand and At Quicksand, the linear plateau was significant when rela-
decreased over time, starting with an average value of 5.9 tive yield was unconstrained (p = 0.0028) and constrained
in 2016 and ending at 5.3 in 2021. We attribute this change (p < 0.0001), with a joint at 10.2 mg kg−1 and 10.0 mg
in pH to the continuous application of N during this study kg−1 , respectively, and pseudo-r2 value of 0.02 and 0.05,
(Schroder et al., 2011). respectively. Across both sites, the linear plateau models were
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8 of 12 REED ET AL.

T A B L E 3 Yield and standard error (SE) for corn yield each year (Mg ha−1 ), the p-value of the model with phosphorus (P) fertilizer treatment
as a dependent variable, and yield as the independent variable.

P treatment Control
Mean (Mg ha−1 ) SE (Mg ha−1 ) Mean (Mg ha−1 ) SE (Mg ha−1 ) p-value
Princeton
2016 10.07 0.16 9.76 0.13 0.0789
2018 7.16 0.11 7.06 0.08 0.7280
2020 3.78 0.11 3.61 0.07 0.1850
Quicksand
2016 11.65 0.10 11.13 0.09 <0.0001
2018 15.49 0.13 14.69 0.11 <0.0001
2020 10.46 0.23 9.94 0.19 0.0001
2021 8.75 0.18 7.97 0.10 <0.0001

Note: Response was significant at α = 0.10.

TA B L E 4 Linear Plateau model parameters for unconstrained and constrained relative yield values across both sites, and with both sites
combined.

Slope Intercept Joint Plateau R2 p-value

Unconstrained relative yield
Princeton – – – – – 0.3770
Quicksand 0.0128 0.83 10.2 0.96 0.02 0.0028
Combined 0.0075 0.88 12.8 0.98 0.02 0.0006
Constrained relative yield (≤1.00)
Princeton 0.0045 0.88 16.5 0.95 0.04 0.0005
Quicksand 0.0124 0.80 10.0 0.92 0.05 <0.0001
Combined 0.0076 0.84 12.7 0.94 0.05 <0.0001

Note: Relative yield was calculated by dividing the yield of the control treatment by the yield of the Phosphorus treatment. Relative yield was either left as calculated
(unconstrained) or constrained at 1.00.

significant for unconstrained (p = 0.0006) and constrained While 96% of our plots had soil test P below the 30 mg kg−1
(p < 0.0001) relative yield, with joints at 12.8 mg kg−1 and critical nutrient threshold where UKY recommends fertilizer
12.7 mg kg−1 , respectively, and pseudo-r2 of 0.02 and 0.05, P (Ritchey & McGrath, 2021), only 51% responded to P
respectively. fertilizer (Table 5), with significant prediction of nutrient
We compared the linear models to determine if the site- response in just two site-years (Princeton in 2020 and
specific models (by site) were more representative of their Quicksand in 2016).
sites compared to models of both sites combined (across At the field scale, the UKY critical level of 30 mg kg−1
sites). Neither the unconstrained (p = 0.3896) or con- effectively identified average P need. We observed average
strained (p = 0.3413) combined model was different from the yield increases in five of seven site-years (Table 3), support-
site-specific models (data not shown). ing UKY’s recommendation for P fertilizer application when
M3P values are below 30 mg kg−1 . However, site-specific
responses varied, with Princeton showing significant yield
4 DISCUSSION response to P in only 1 year (2016), while Quicksand showed
positive yield response in all years. We attributed the lack
Our objective was to evaluate soil test P effectiveness for of positive response at Princeton to poor crop stands and
variable rate P management within fields and to understand adverse growing conditions in 2018 and 2022, where yields
if soil test critical levels hold at the subfield level. Our results lagged state averages. In contrast, Quicksand provided yields
indicated that within fields, crop response was uncorrelated to comparable to state averages throughout the study period
soil test P at M3P below critical concentrations for Kentucky. (USDA-NASS, 2024). While only 4% of our main plots had
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
REED ET AL. 9 of 12

F I G U R E 6 Linear plateau models for unconstrained and constrained relative yield values by and across sites. Relative yield was calculated by
dividing the yield of the control treatment by the yield of the phosphorus treatment. Relative yield was either left as calculated (unconstrained) or
constrained at 1.0.

T A B L E 5 Summary Statistics of main plot Mehlich 3 extractable phosphorus (M3P) content of both plots that had a response and those that did
not respond to the addition of phosphorus containing fertilizer across both sites and all years.

No response Response
Min M3P Max M3P Mean M3P Min M3P Max M3P Mean M3P
n (mg kg−1 ) (mg kg−1 ) (mg kg−1 ) n (mg kg−1 ) (mg kg−1 ) (mg kg−1 ) p-value
Princeton
2016 49 11 50 22 39 10 52 23 0.4290
2018 100 6 42 14 45 6 35 13 0.1740
2020 67 1 23 10 66 1 21 8 0.0045
Quicksand
2016 58 7 36 15 61 6 31 12 0.0038
2018 77 6 33 12 88 6 47 11 0.4130
2020 38 4 26 11 87 3 31 10 0.2469
2021 60 3 20 10 76 3 31 12 0.1410
Combined <0.0001

Note: We determined that a main plot was responsive to P if the yield of the P treatment exceeded the mean yield plus one standard deviation of the unfertilized control
subplots. For main plots where standard deviation was not available due to missing data, we used the standard deviation of the control treatments at that site-year in place
of the individual plot unfertilized standard deviation. Generalized linear model was used to determine if M3P was a predictor of yield response (yes/no). Response was
significant at an α = 0.10.

M3P >30 mg kg−1 , with an average RY of 0.96 they created a been documented in other corn and soybean studies (Fulford
definitive plateau. This aligns with the fertilizer recommenda- & Culman, 2018; Reed et al., 2022).
tion support tool project, which reported a plateau RY of 92% Linear plateau models revealed important insights about
at a Mehlich 3 P of 43 mg kg−1 based on 177 corn response spatial variability in P response (Figure 6). There were no sig-
trials across 17 states using a 15 cm soil sample (Buol et al., nificant differences between site-specific and combined-site
2024). This variable response pattern at the field level has models, with joint points ranging from 10.0 to 16.5 mg kg−1
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10 of 12 REED ET AL.

(Table 4). The poor pseudo-r2 values (0.02–0.05) indicated the critical P threshold recommended for fertilizer appli-
minimal differences in model fits across datasets. While Dodd cation, only 51% responded to P inputs, underscoring the
and Mallarino (2005) noted different joint points between limitations of relying solely on soil test P for precision
sites, our spatially dense dataset specifically examined P agriculture. The inability of soil test P alone to accurately pre-
response within fields, revealing limitations in traditional dict yield response highlights the complexity of P dynamics
approaches. Though we focused on linear plateau models, within fields and the influence of other contributing factors.
which provide the basis for UKY recommendations, other Future precision agriculture models require innovative exper-
modeling approaches (Slaton et al., 2024) would not alter our imental designs to capture the spatial variability of yield
fundamental finding: yield response to P exhibits substantial response to P as a function of additional soil properties,
spatial and temporal variability that soil test correlation alone crop characteristics, and environmental conditions. Our novel
cannot precisely predict. approach of collecting spatially dense yield response data
Our data reveal that conventional experimental designs and correlated against soil P concentrations provides a frame-
statistical treatment of soil test correlation data, used to make work for developing more precise fertilizer recommendations
state-level recommendations (Lyons et al., 2021), mask sub- that optimize crop yields while minimizing environmental
field variability that should be accounted for in variable rate impacts.
fertilizer management. Notably, our study showed RY val-
ues commonly above 1.0, suggesting negative responses to AU T H O R C O N T R I B U T I O N S
P fertilizer. While negative impacts from P fertilizer appli- Vaughn Reed: Formal analysis; validation; visualization;
cation are presumed rare in the literature, this observation writing—original draft; writing—review and editing. Jenni
raises important questions about whether constraining RY Fridgen: Formal analysis; investigation; writing—review
values in models might mask actual spatial variability of nutri- and editing. Bronc Finch: Investigation; writing—review
ent response. As noted previously, the magnitude of positive and editing. John Spargo: Formal analysis; methodology;
response at low soil test P ranges outweighed the negative writing—review and editing. Josh McGrath: Conceptu-
responses, so that the mean effect of P application was pos- alization; data curation; formal analysis; funding acqui-
itive. Overall, the probability and magnitude of P response sition; investigation; methodology; project administration;
decreased as soil test increased, aligning with the traditional resources; supervision; validation; writing—review and edit-
understanding of soil test correlation. ing. James M. Bowen: Conceptualization; data curation;
Traditional variable rate P management assumed that soil investigation; methodology; writing—original draft. Gene
test correlation could be applied at the subfield level with Hahn: Methodology; supervision. Douglas Smith: Fund-
high-density sampling. However, effective representation of ing acquisition; investigation; supervision. Edwin Ritchey:
soil nutrient variability requires prohibitively expensive sam- Resources; supervision.
pling grids <30 m (Lauzon et al., 2005). Our findings,
supported by recent research (Culman et al., 2023; Reed AC K N OW L E D G M E N T S
et al., 2021, 2022), suggest that even high-density soil P This project was supported through the US Department
mapping cannot overcome the inherent variability in subfield of Agriculture’s Conservation Effects Assessment Project
response. Future research should explore incorporating mech- (CEAP), a multi-agency effort led by the Natural Resources
anistic factors such as soil texture, climate zones, and crop Conservation Service (NRCS) to quantify the effects of vol-
production history (Beneduzzi et al., 2022; Jordan-Meille untary conservation and strengthen data-driven management
et al., 2012; Peltovuori, 1999; Ramamurthy et al., 2009) or decisions across the nation’s private lands. We would like
develop econometric approaches using probabilistic models to acknowledge the contributions of James “Jimmy” Michael
to account for stochastic variation and hedge against economic Bowen, a PhD student who was an integral part of this project.
loss. These strategies could better support precision fertilizer Jimmy had completed a significant portion of the research in
management while acknowledging inherent response variabil- pursuit of his degree prior to his passing. We also would like
ity, potentially improving both economic and environmental to acknowledge the support of the International Plant Nutri-
outcomes (Zhang et al., 2024). tion Institute (IPNI) Phosphorus Fellowship Program, which
was instrumental in making this research possible.

5 CONCLUSION C O N F L I C T O F I N T E R E ST STAT E M E N T
The authors declare no conflicts of interest.
This study demonstrated that soil test P, while effective at
predicting average field-level nutrient responses, lacks the ORCID
precision required for sub-field, variable rate fertilizer man- Vaughn Reed https://orcid.org/0000-0001-6529-4570
agement. Although 96% of plots in this study fell below Bronc Finch https://orcid.org/0000-0001-9571-2438
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
REED ET AL. 11 of 12

John Spargo https://orcid.org/0000-0002-7400-4739 specific management zones. Soil Science Society of America Journal,
Josh McGrath https://orcid.org/0000-0002-1212-8795 69(5), 1572–1579. https://doi.org/10.2136/sssaj2005.0062
Osmond, D. L., Kleinman, P. J. A., Coale, F., Nelson, N. O., Bolster,
C. H., & McGrath, J. (2024). A short history of the phosphorus
REFERENCES
index and Andrew Sharpley’s contributions from inception through
Beneduzzi, H. M., Souza, E. G., Moreira, W. K., Sobjak, R., Bazzi,
development and implementation. Journal of Environmental Quality.
C. L., & Rodrigues, M. (2022). Fertilizer recommendation methods
https://doi.org/10.1002/jeq2.20535
for precision agriculture—A systematic literature study. Engen-
Pearce, A. W., Slaton, N. A., Lyons, S. E., Bolster, C. H., Bruulsema,
haria Agrícola, 42, e20210185. https://doi.org/10.1590/1809-4430-
T. W., Grove, J. H., Jones, J. D., McGrath, J. M., Miguez, F. E.,
eng.agric.v42n1e20210185/2022
Nelson, N. O., Osmond, D. L., Parvej, M. R., Pena-Yewtukhiw, E.
Buol, G., Osmond, D., Slaton, N., Spargo, J., Lyons, S. E., Pearce, A.,
M., & Spargo, J. T. (2022). Defining relative yield for soil test corre-
Uthman, Q., Yost, M., & Kaiser, D. E. (2024). Fertilizer recom-
lation and calibration trials in the fertilizer recommendation support
mendation support tool (Version V 1.0 0.0) [Computer software].
tool. Soil Science Society of America Journal, 86(5), 1338–1353.
https://frst.scinet.usda.gov/tool
https://doi.org/10.1002/saj2.20450
Culman, S., Fulford, A., LaBarge, G., Watters, H., Lindsey, L. E.,
Peltovuori, T. (1999). Precision of commercial soil testing practice for
Dorrance, A., & Deiss, L. (2023). Probability of crop response to
phosphorus fertilizer recommendations in Finland. https://doi.org/10.
phosphorus and potassium fertilizer: Lessons from 45 years of Ohio
23986/afsci.5631
trials. Soil Science Society of America Journal, 87(5), 1207–1220.
Penn, C. J., & Camberato, J. J. (2019). A critical review on soil chemical
https://doi.org/10.1002/saj2.20564
processes that control how soil pH affects phosphorus availabil-
Dari, B., Rogers, C. W., Leytem, A. B., & Schroeder, K. L. (2019). Eval-
ity to plants. Agriculture, 9(6), Article 120. https://doi.org/10.3390/
uation of soil test phosphorus extractants in Idaho soils. Soil Science
agriculture9060120
Society of America Journal, 83(3), 817–824. https://doi.org/10.2136/
Pierpaoli, E., Carli, G., Pignatti, E., & Canavari, M. (2013). Drivers of
sssaj2018.08.0314
precision agriculture technologies adoption: A literature review. Pro-
Dodd, J. R., & Mallarino, A. P. (2005). Soil-test phosphorus and crop
cedia Technology, 8, 61–69. https://doi.org/10.1016/j.protcy.2013.11.
grain yield responses to long-term phosphorus fertilization for corn-
010
soybean rotations. Soil Science Society of America Journal, 69(4),
R Core Team. (2022). R: A language and environment for statistical
1118–1128. https://doi.org/10.2136/sssaj2004.0279
computing. R Foundation for Statistical Computing. https://www.R-
Fulford, A. M., & Culman, S. W. (2018). Over-fertilization does not
project.org/
build soil test phosphorus and potassium in Ohio. Agronomy Journal,
Ramamurthy, V., Naidu, L. G. K., Kumar, S. C. R., Srinivas, S., &
110(1), 56–65. https://doi.org/10.2134/agronj2016.12.0701
Hegde, R. (2009). Soil-based fertilizer recommendations for precision
Hansel, F. D., Amado, T. J. C., Ruiz Diaz, D. A., Rosso, L. H.
farming. Current Science, 97(5), 641–647.
M., Nicoloso, F. T., & Schorr, M. (2017). Phosphorus fertilizer
Reed, V., Souza, J. L. B., Lofton, J., & Arnall, B. (2022). On farm
placement and tillage affect soybean root growth and drought tol-
evaluation of preplant soil test P and K in double crop soybeans.
erance. Agronomy Journal, 109(6), 2936–2944. https://doi.org/10.
Agrosystems, Geosciences & Environment, 5(4), e20307. https://doi.
2134/agronj2017.04.0202
org/10.1002/agg2.20307
Hopkins, B. G., & Hansen, N. C. (2019). Phosphorus management in
Reed, V., Watkins, P., Souza, J., & Arnall, B. (2021). Evaluation of incor-
high-yield systems. Journal of Environmental Quality, 48(5), 1265–
porated phosphorus fertilizer recommendations on no-till managed
1280. https://doi.org/10.2134/jeq2019.03.0130
winter wheat. Crop, Forage & Turfgrass Management, 7(2), e20133.
Howard, D. D., Essington, M. E., & Tyler, D. D. (1999). Vertical
https://doi.org/10.1002/cft2.20133
phosphorus and potassium stratification in no-till cotton soils. Agron-
Ritchey, E., & McGrath, J. M. (2021). AGR-1: Lime and fertilizer recom-
omy Journal, 91(2), 266–269. https://doi.org/10.2134/agronj1999.
mendations. University of Kentucky Extension Service. http://www2.
00021962009100020014x
ca.uky.edu/agcomm/pubs/AGR/AGR1/AGR1.pdf
Jordan-Meille, L., Rubæk, G. H., Ehlert, P., Genot, V., Hofman, G.,
Schroder, J. L., Zhang, H., Girma, K., Raun, W. R., Penn, C. J., & Payton,
Goulding, K., Recknagel, J., Provolo, G., & Barraclough, P. (2012).
M. E. (2011). Soil acidification from long-term use of nitrogen fertil-
An overview of fertilizer-P recommendations in Europe: Soil testing,
izers on winter wheat. Soil Science Society of America Journal, 75(3),
calibration and fertilizer recommendations. Soil Use and Man-
957–964. https://doi.org/10.2136/sssaj2010.0187
agement, 28(4), 419–435. https://doi.org/10.1111/j.1475-2743.2012.
Sikora, F. J., & Moore, K. P. (Eds.). (2014). Soil test meth-
00453.x
ods from the southeastern United States (Southern Cooperative
Lauzon, J. D., O’Halloran, I. P., Fallow, D. J., von Bertoldi, A. P., &
Series Bulletin No. 419). Southern Extension and Research Activ-
Aspinall, D. (2005). Spatial variability of soil test phosphorus, potas-
ity Information Exchange Group 6. http://aesl.ces.uga.edu/sera6/
sium, and pH of Ontario soils. Agronomy Journal, 97(2), 524–532.
MethodsManualFinalSERA6.pdf
https://doi.org/10.2134/agronj2005.0524
Slaton, N., Pearce, A., Gatiboni, L., Osmond, D., Bolster, C., Miquez,
Lyons, S. E., Arthur, D. K., Slaton, N. A., Pearce, A. W., Spargo, J.
F., Clark, J., Dhillon, J., Farmaha, B., & Kaiser, D. (2024). Models
T., Osmond, D. L., & Kleinman, P. J. A. (2021). Development of a
and sufficiency interpretation for estimating critical soil test values
soil test correlation and calibration database for the USA. Agricul-
for the fertilizer recommendation support tool. Soil Science Society
tural & Environmental Letters, 6(4), e20058. https://doi.org/10.1002/
of America Journal, 88(4), 1419–1437. https://doi.org/10.1002/saj2.
ael2.20058
20704
Mzuku, M., Khosla, R., Reich, R., Inman, D., Smith, F., & MacDonald,
Slaton, N. A., Lyons, S. E., Osmond, D. L., Brouder, S. M., Culman,
L. (2005). Spatial variability of measured soil properties across site-
S. W., Drescher, G., Gatiboni, L. C., Hoben, J., Kleinman, P. J. A.,
 14350645, 2025, 1, Downloaded from https://acsess.onlinelibrary.wiley.com/doi/10.1002/agj2.70028 by Nolan Mullican - Mississippi State University , Wiley Online Library on [27/02/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12 of 12 REED ET AL.

McGrath, J. M., Miller, R. O., Pearce, A., Shober, A. L., Spargo, J. T., Zhang, J., Sleutel, S., & Mouazen, A. M. (2024). Phosphorus-based
& Volenec, J. J. (2022). Minimum dataset and metadata guidelines for variable rate manure application in wheat and barley. Precision
soil-test correlation and calibration research. Soil Science Society of Agriculture, 25(3), 1714–1730. https://doi.org/10.1007/s11119-024-
America Journal, 86(1), 19–33. https://doi.org/10.1002/saj2.20338 10131-2
Solie, J., Raun, W., & Stone, M. (1999). Submeter spatial variability
of selected soil and bermudagrass production variables. Soil Science
Society of America Journal, 63(6), 1724–1733. https://doi.org/10. How to cite this article: Reed, V., Fridgen, J., Finch,
2136/sssaj1999.6361724x B., Spargo, J., McGrath, J., Bowen, J. M., Hahn, G.,
Souza, J. L. B. (2020). Stratification of soil characteristics in long Smith, D., & Ritchey, E. (2025). Soil test phosphorus
term winter wheat fertility studies after 9 years of no-till manage- predicts field-level but not subfield-level corn yield
ment in oklahoma. ASA-CSSA-SSSA International Annual Meeting,
response. Agronomy Journal, 117, e70028.
Phoenix, AZ.
USDA-NASS. (2024). NASS—Quick stats. United States—National
https://doi.org/10.1002/agj2.70028
Agricultural Statistics Service. https://quickstats.nass.usda.gov/