Using the Haney Soil Test to Predict Nitrogen Requirements in Winter Wheat (Triticum aestivum L.)

Abstract

Managing nitrogen (N) is one of the of the biggest challenges in achieving environmental and economic sustainability in the agroecosystem. As N fertilizer prices have increased significantly, farmers are considering a revised N recommendation to optimize crop production, while addressing negative environmental impacts of excess N in water bodies. This study analyzes the accuracy of using the Haney Soil Test (HST) to predict the N requirement (HSTNR) of winter wheat (Triticum aestivum L.) in a semi-arid climate. The accuracy of the HST to predict the economically optimum N rate (EONR) was dependent on in-season precipitation. In drought conditions, the HSTNR was 33 kg N ha⁻¹ lower on average than the EONR. Conversely, in wetter years, the HSTNR was 35 kg N ha⁻¹ higher than the EONR. Net return was approximately USD 19 ha⁻¹ lower than that with the EONR under both precipitation scenarios. Similar differences were found for protein content. There was a strong correlation between soil respiration and the soil health calculation, within the HST, and the difference between the net return on yield from the HSTNR and the EONR yield. These indicators may serve as useful metrics for formulating soil health-based N recommendations in winter wheat. However, in drought-prone areas, the HSTNR may significantly underpredict the EONR in many years due to an overestimation of N mineralization.

1. Introduction

Mitigating agricultural nitrogen (N) pollution is one of the major environmental concerns in the twenty-first century [1,2]. Additionally, as N fertilizer prices increase, growers are considering a revision of their current N recommendations to use an optimum N rate to enhance N use efficiency [3]. N fertilizer recommendations must bridge the shortfall between N provided by the soil and that required by the plant. To be most effective in improving current N recommendations, new methodologies need to accurately incorporate N provided by the soil and optimize economic profits. Some current N fertilizer recommendations incorporate economics to calculate an economically optimal N rate (EONR) by including the cost of fertilizer, the price of the commodity, and the response curve of the crop to N additions into a rate recommendation—independent of soil conditions [4,5].

Conversely, the Haney Soil Test (HST) was developed to include a suite of lab-based methods to predict the amount of N supplied to the crop from the soil, which could improve the accuracy of N fertilizer recommendations [6,7,8]. Traditional soil testing uses potassium chloride (KCl) to extract plant-available N from the soil, while the HST utilizes an extractant comprising organic acids designed to mimic the mechanisms used by plant roots to acquire nutrients at a solution pH similar to that observed in the soil [8]. Additionally, the HST approach assesses water-extractable organic carbon and organic N in the soil along with soil respiration as a 1-day CO₂ evolution to estimate the microbial contribution to plant-available N [7,9,10].

Initial evaluations of the HST have primarily focused on the southern and central United States, with a particular focus on corn (Zea mays L.) production [11,12]. It is less clear how well the HST predicts N supply in more northern climates with small grains. Wheat (Triticum aestivum L.) differs from many other commodity crops in that protein content is measured along with yield. Because wheat protein is well correlated with N fertility, one concern from producers is that reducing N fertilizer will reduce protein, even if yields remain adequate. The objective of the current study was to compare two approaches in estimating the N requirement of wheat: an economics-based approach that does not consider soil N mineralization with the HST method that specifically measures soil N supply. Thus, our study evaluated the accuracy of the HST for N recommendations in winter wheat in a semi-arid climate where N limitations often reduce yield and protein content due to slow rates of mineralization.

2. Materials and Methods

2.1. Experimental Design and Setup

This study took place over three growing seasons (2017–2019) at three locations (Sturgis, Vivian and Wall) in western South Dakota. However, the field sites where experimental plots were laid out were different for different years, but had similar soil types and were managed similarly. Initial soil test information is given in

Table 1. Initial values for soil organic matter (SOM), pH, electrical conductivity (EC), organic nitrogen released (ON release), total Haney available nitrogen (N), and soil health calculations (SHC) using Haney Soil Test protocols.

Site	SOM %	pH	EC	ON Release 0–15 cm	ON Release 15–60 cm	Total Haney Available N	SHC 0–15 cm	SHC 15–60 cm
			dS m−1	Kg N ha−1
Sturgis 2017	3.1	6.8	0.5	22	-	33	16.5	-
Vivian 2017	4.4	7.3	0.8	7	-	33	5.5	-
Wall 2017	1.8	6.4	0.3	21	-	63	11.9	-
Sturgis 2018	3.6	6.1	0.3	30	25	67	15.4	10.7
Wall 2018	2.7	6.9	1.0	50	27	92	15.6	7.9
Vivian 2019	2.6	6.5	0.3	25	22	56	7.6	6.1
Wall 2019	5.2	7.8	0.7	36	35	116	22.4	20.4

. Two site-years (Vivian, 2018 and Sturgis, 2019) were omitted from the study due to unresponsiveness to N and flooding, respectively. As excess precipitation coupled with soil texture created an anaerobic condition, it impacted the N dynamics; therefore, the research dataset could not be used to maintain a logical interpretation of the overall research outcomes. Hard red winter wheat (HRW) was planted in the Fall on 25 cm row spacing with a no-till grain drill (Model 750, John Deere Co., Moline, IL, USA) at a population of 297 pure live seeds m⁻². In 2017, urea granules were broadcast at rates of 0, 45, 90, 134, 180 and 224 kg N ha⁻¹ at planting. In 2018 and 2019, urea was applied at the ‘Feekes 4’ growth stage of wheat, also known as ‘green-up’ when active tiller formation occurs [13]. These differences in N application dates are assumed to be negligible, as previous research suggests insignificant differences in wheat production due to early-season application timing in semi-arid climates [14].

All sites used the same randomized complete block design with four replications. Plots consisted of six planted rows with a total area of 14 m². At harvest maturity, the front and back of each plot were trimmed back and only the middle four rows were harvested to avoid edge effects (total harvested area = 11.6 m²). Plots were harvested with a small plot combine (Wintersteiger, Salt Lake City, UT, USA), and yield was adjusted to 130 g H₂O kg-grain⁻¹. A subsample from each replication was taken for further protein analysis using whole grain near-infrared transmittance (Foss, Eden Prairie, MN, USA). Each protein value was a composite of the measured light transmittance through the grain from ten sub-samples.

2.2. Soil Sampling and Haney N Recommendations

Soil samples for N analysis were gathered in early spring prior to urea application. Each sample consisted of a composite of 8–12 samples per plot. In 2017, N recommendations were based on a 0–15 cm sample only per HST recommendations. In 2018 and 2019, samples were taken from 0–15 cm and 15–60 cm depths to align more closely with standard N testing for wheat in South Dakota [15].

Haney soil test protocols were conducted at the USDA-ARS Grassland Soil and Water Research Laboratory on air-dried samples previously passed through a 2-mm sieve. All analyses in the HST protocol were based on previously described methods, which together were combined to create a N recommendation (HSTNR) based on estimated plant-available N [6,7,8,16,17,18]. The final HSTNR was determined using guidelines established by the South Dakota State Experiment Station [15]:

Haney Soil Test Fertilizer Recommendation (HSTNR) = 2.5 × YG − HSTN

where YG is the yield goal in bu ac⁻¹ (where 1 bushel of wheat weighs 60 lbs.; therefore, 1 bu acre⁻¹ is equal to 67.25 kg ha⁻¹; we converted YG to kg ha⁻¹), ‘2.5’ is a numeric factor, derived from N rate studies in South Dakota [15]; HSTN is the estimated HST plant-available N (kg ha⁻¹) from either a 0–15 cm or 0–60 cm soil sample.

To determine how well the HSTNR approximates the EONR, the YG was determined as the yield corresponding to the EONR of each site and year. The maximum return to N (MRTN) approach (the N rate that maximizes the economic return to N application) was used to determine the EONR and corresponding yield [4,19,20]. This approach predicts N rates based on replicated N response trials, independent of starting soil N values. However, the maximum grain accumulation is the same with or without accounting for soil N, hence the validity of using this value as the YG. Based on the similarity of fit statistics but significantly different join point, both a quadratic plateau and a linear plateau curve were used to fit the yield response for each N trial conducted (

Table 2. Regression statistics for the linear plateau (LP) and quadratic plateau (QP) models and plateau N rate and corresponding grain yield at the hinge point, where R² is the coefficient of determination, RMSE is root mean square error, and AIC is Akaike information criterion.

Location	Year	LP R2	QP R2	LP RMSE	QP RMSE	LP AIC	QP AIC	LP Plateau N	QP Plateau N	LP Plateau Grain Yield	QP Plateau Grain Yield
								kg ha−1
Sturgis	2017	0.81	0.80	241	248	339	341	134	204	2481	2523
Sturgis	2018	0.60	0.60	424	425	366	367	72	109	3481	3487
Vivian	2017	0.81	0.81	322	324	353	354	101	155	3531	3558
Vivian	2019	0.80	0.81	218	212	335	333	106	159	4304	4318
Wall	2017	0.50	0.51	495	495	375	374	95	120	3278	3267
Wall	2018	0.70	0.71	414	410	336	335	111	162	5446	5458
Wall	2019	0.87	0.89	402	372	364	360	101	136	4641	4624
Average		0.73	0.73	359	355	353	352	103	149	3880	3891

). The linear plateau model is defined by Equations (l) and (2):

$$Y=a+bX if X<J$$

(1)

$$Y=P if X\ge J$$

(2)

where Y is the yield of grain (kg ha⁻¹) and X is the N application rate (kg ha⁻¹); a (intercept), b (linear coefficient), J (join point, occurring at the intersection of the linear and the plateau lines), and P (plateau yield) are constants obtained by fitting the model to the data. The quadratic-plus-plateau model is defined by Equations (3) and (4):

$$Y=a+bX+c{X}^2 if X<J$$

(3)

$$Y=P if X\ge J$$

(4)

where Y is the yield of grain (kg ha⁻¹) and X is the N application rate (kg ha⁻¹); a (intercept), b (linear coefficient), c (quadratic coefficient), J (join point, occurring at the intersection of the quadratic and the plateau lines), and P (plateau yield) are constants obtained by fitting the model to the data [21]. The join point (J) is considered to be the point at which increasing the fertilizer rate is no longer effective at increasing yield. The plateau (P) is the value at which yield is maximized for the site.

Using the equation derived from fitting each model (i.e., site) (Figure 1), the net return can be calculated and plotted as the increase in yield multiplied by the grain price at a given N rate, minus the cost of that same amount of N. In this paper, we used a ratio of 5.55 for cost of fertilizer N: price of wheat grain (equivalent to USD 1.11 kg N⁻¹ (USD 0.50 lb N⁻¹) and USD 0.202 kg grain⁻¹ (USD 5.50 bu grain⁻¹)). The EONR is the N rate at which this return is maximized.

All statistical analyses were conducted in the R statistical package (R Core Development Team, 2014). Analysis examined both differences between the HSTNR and EONR using either the plant-available N estimated from 0–15 cm soil depth or as the total plant-available N from the 0–60 cm soil depth where available (2018 and 2019).

3. Results and Discussion

Precipitation during the three growing seasons strongly influenced the differences between HSTNR and MRTN (Figure 2 and Figure 3). The 2017 growing season experienced severe drought, receiving 60–70% of the long-term average precipitation (https://climate.sdstate.edu/: accessed on 1 July 2021). During this year, the HSTNR was underestimated by an average of 35 kg ha⁻¹ for the quadratic plateau yield curve and 30 kg ha⁻¹ for the linear plateau estimation using a 0–15 cm soil sample. Conversely, in wetter years (2018 and 2019, https://climate.sdstate.edu/: accessed on 1 June 2021), the HSTNR was on average 29 kg ha⁻¹ and 41 kg ha⁻¹ higher than the MRTN for the quadratic plateau and linear plateau curves, respectively (

Table 3. Estimated nitrogen rate (kg ha⁻¹) for the linear plateau (LP) and quadratic plateau (QP) models at the estimated Haney Soil Test (HST) and MRTN values.

Site	Estimated Available N (0–15 cm)	Estimated Available N (0–60 cm)	HSTNR QP (0–15 cm)	HSTNR LP (0–15 cm)	HSTNR QP (0–60 cm)	HSTNR LP (0–60 cm)	QP MRTN	LP MRTN
	kg N ha−1
Sturgis 2017	37	-	58	66	-	-	122	134
Vivian 2017	37	-	106	109	-	-	119	101
Wall 2017	71	-	61	65	-	-	91	95
Sturgis 2018	39	75	103	105	67	69	87	72
Wall 2018	31	63	191	195	159	163	123	111
Vivian 2019	62	130	110	117	42	49	101	106
Wall 2019	27	41	163	166	149	152	118	101
Average			113	118	104	108	109	103

). It is important to note that the HST is a lab-based method that measures the CO₂ flux following soil rewetting to predict the potential availability of N to the plant through mineralization [11]. This is due in large part to the strong correlation between soil respiration and water-soluble organic N [7]. Further, soils in more arid environments tend to exhibit a lower CO₂ burst upon rewetting, which is presumed to be due to a higher frequency of drying/rewetting cycles and an increasing trend of available carbon depletion [22]. However, where the rewetting cycle is limited (e.g., drought), so too is respiration, which presumes a decreased rate of N mineralization and an overprediction of N available to the plant.

Moreover, using a 0–60 cm soil sample will always decrease the HSTNR over a 0–15 cm soil sample because it incorporates a larger soil volume and accounts for more soil N. This means that differences between the HSTNR and MRTN will likely be exacerbated during dry years but may be closer to the optimum in wetter years/climates. Overall, incorrect estimation of N mineralization can be attributed to its dependency on climatic variables, specifically rainfall [23,24,25].

Because the HSTNR varied by weather, wheat yields were similarly affected. During the dry 2017 growing season, the estimated HST yield was 276 kg ha⁻¹ lower than the MRTN using a quadratic plateau curve and was 361 kg ha⁻¹ lower when a linear plateau curve was used (

Table 4. Estimated grain yield (kg ha⁻¹) for the linear plateau (LP) and quadratic plateau (QP) models at the estimated Haney soil test (HST) and MRTN values.

Site	HST QP Yield (0–15 cm)	HST LP Yield (0–15 cm)	HST QP Yield (0–60 cm)	HST LP Yield (0–60 cm)	QP MRTN Yield	LP MRTN Yield
	kg grain ha−1
Sturgis 2017	1806	1814	-	-	2294	2481
Vivian 2017	3369	3531	-	-	3454	3531
Wall 2017	2932	2863	-	-	3186	3278
Sturgis 2018	3481	3481	3268	3421	3424	3481
Wall 2018	5458	5456	5450	5456	5352	5446
Vivian 2019	4204	4304	3672	3688	4158	4304
Wall 2019	4625	4641	4625	4641	4572	4641
Average	3696	3727	4254	4302	3777	3880

). This was largely due to two of the three sites having HSTNR values significantly lower than the MRTN rate (Table 3). Ironically, the Vivian site’s 2017 HSTNR was very close to the MRTN rate, but this was due to the low soil health score from the Haney test, which resulted in a low predicted N mineralization (hence, higher predicted HSTNR). During more conducive growing conditions, the HST yield using a linear plateau estimation generally fell beyond the curve join point, resulting in N being applied without returning an increase in grain yield. Using a quadratic plateau also resulted in yields beyond the MRTN yield but to a lesser extent, averaging 66 kg ha⁻¹ more than the MRTN (Table 4).

From an economic perspective, the quadratic plateau estimation of the MRTN was always lower than the linear plateau model (

Table 5. Estimated returns (USD ha⁻¹) for the linear plateau (LP) and quadratic plateau (QP) model parameters at the estimated Haney soil test (HST) and MRTN values.

Site	HSTN QP Return (0–15 cm)	HSTN LP Return (0–15 cm)	HSTN QP Return (0–60 cm)	HSTN LP Return (0–60 cm)	QP MRTN Return	LP MRTN Return
	USD ha−1
Sturgis 2017	300.38	293.14	-	-	328.01	352.42
Vivian 2017	562.87	592.27	-	-	565.59	601.15
Wall 2017	524.49	506.13	-	-	542.57	556.71
Sturgis 2018	588.85	586.61	585.86	614.44	595.05	623.24
Wall 2018	890.51	885.66	933.28	921.18	944.54	978.90
Vivian 2019	727.03	739.54	695.04	690.6	727.84	751.75
Wall 2019	753.26	768.8	753.22	768.76	792.66	825.37
Average	621.06	624.59	741.85	748.75	642.32	669.93

). By definition, the MRTN value will always equal the plateau/join point, whereas with the quadratic plateau model, the MRTN value is generally significantly lower than the plateau join point. As a result, net return is more volatile in the linear plateau estimation and, in general, the net HSTN return was closer to the MRTN in the quadratic model. On average, the quadratic plateau model HSTN return was USD 21.27 ha⁻¹ less than the MRTN, whereas the linear plateau model HSTN was USD 45.34 less than the MRTN. These differences were roughly 97% and 93% of the MRTN for the quadratic plateau and linear plateau models, respectively (Table 5).

Components of the HST have been shown to be moderately correlated to the EONR of corn [12]. While the data are limited, our study supports this finding in wheat as a strong correlation was found in the difference between net return from the HSTN and the EONR return and either the soil health calculation (SHC) or the 1-day CO₂ burst (Figure 4). The SHC is a relative measure of soil health, which incorporates the 1-day CO₂ burst (hence autocorrelation) along with measures of soil organic C and N. While not used explicitly in the HSTNR, it appears to offer a potential guide for estimating N recommendations along with CO₂ evolution. The negative correlation in average or above average precipitation growing seasons suggests that a higher SHC or CO₂ burst value indicates that the HSTNR was a more accurate approximation of the EONR in this study. Additionally, the positive correlation in the drier year corroborates our inference that the HST is overestimating the N supplying abilities of the soil with the implicit assumption that a high SHC (i.e., a healthier soil) equates to more N mineralization and a lower HSTNR. Another report from South Dakota also recommended CO₂ burst (soil respiration) as a potential tool for N recommendation in corn [26,27]. Further study should be directed toward the influence of rainfall on the estimation of the HSTNR to ensure that the effect is directly related to rainfall and not confounded by other site-specific factors such as soil texture and land use history.

The flush of CO₂ is a measure of microbial biomass and well-correlated with N mineralization, but this is a simulated measure under idealized conditions [17]. Our results suggest that the efficacy of the ‘CO₂ burst’ would be maximized by incorporating as much in-season weather as possible. Hence, the later the soil test is taken into the growing season, the closer it is likely to approximate the EONR, which may necessitate a split N application in wheat; similar reports are available for corn [26,27,28].

Likewise, protein content is a critical measure of wheat quality and is positively correlated with timing and concentration of plant-available N [29,30]. Hence, protein content at the HSTNR was not significantly different between the quadratic plateau and linear plateau models (

Table 6. Estimated parameterized percent protein values based on the regression of applied fertilizer N on protein content for each site-year.

Site-year	Regression Equation	R2	HSTN QP	HSTN LP	HSTN QP	HSTN LP	EONR QP	EONR LP
			0–15 cm		0–60 cm
Sturgis 2017	y = 0.016x + 11.26	0.80	12.2	12.3	-	-	13.2	13.4
Vivian 2017	y = 0.017x + 11.15	0.78	13.0	13.0	-	-	13.2	12.9
Wall 2017	y = 0.024x + 9.55	0.87	11.0	11.1	-	-	11.7	11.8
Sturgis 2018	y = 0.013x + 12.13	0.61	13.5	13.5	13.0	13.0	13.3	13.1
Wall 2018	y = 0.021x + 9.05	0.92	13.1	13.1	12.4	12.5	11.6	11.4
Vivian 2019	y = 0.010x + 11.99	0.89	13.1	13.2	12.4	12.5	13.0	13.1
Wall 2019	y = 0.011x + 10.49	0.65	12.3	12.3	12.1	12.2	11.8	11.6
Average	y = 0.016x + 10.80	0.79	12.6	12.7	12.5	12.5	12.5	12.4

). During the drier 2017 growing season, the HSTNR estimated protein trended much lower, which could be problematic in years where protein discounts and premiums are in effect. Winter wheat typically requires a protein content of 12%. Where nitrogen mineralization does not meet expectations (i.e., dry growing conditions), a protein shortfall is more likely. Therefore, a split application of N towards reproductive growth stages can provide additional benefit by improving wheat quality; a recent report indicated similar findings in wheat [14,31].

4. Conclusions

Devising efficient N recommendations is difficult because historically it has been largely based on available N (generally nitrate N) in soil before/at the time of planting and does not consider N mineralization during crop growth. The HST attempts to account for N mineralization in its recommended N rate, which offers the prospect of reducing overall N rates. The SHC and 1-day CO₂ burst provide useful indicators of the potential for the soil to supply N to the plant. However, in-season precipitation plays a critical role in the efficacy of this test, especially in dryland ecosystems, where plant-available water is primarily dependent on in-season precipitation, ultimately influencing N mineralization. In our study, the HST appeared to underestimate N mineralization under dry conditions. Since N mineralization is a microbially mediated process and water limitations severely reduce microbial metabolism and growth, drought effectively ‘penalized’ the soils with higher mineralization potential (i.e., ‘healthy soils’ with higher SOM, mineralizable N, soil protein content or microbial activities), where that potential was not realized in-season due to an assumed limitation on N mineralization during critical plant growth stages. In semi-arid climates, where adequate protein concentration is of concern, a 0–15 cm soil sample for HSTNR, rather than a 0–60 cm sample, likely approximates the EONR more closely with adequate protein in drought years or when N is applied as a single application. However, under wetter conditions, a 0–60 cm soil sample may perform similarly, particularly where a split application is utilized for protein content.

Designing N recommendations that incorporate N mineralization during crop growth is critical to improving N recommendations overall. Under dryland conditions in particular, attaining accuracy in estimating N mineralization is difficult. However, split N application provides an opportunity to incorporate more weather information during the wheat growing season to better optimize N use efficiency. Our study provides valuable information regarding the impact of in-season precipitation on HST-based N recommendation that necessitates a revision of the parameters used in the estimation of model parameters. In spite of other limitations in the scope of this study due to limited numbers of years and locations, it indicates an important knowledge gap to improve N recommendation. Future studies should incorporate more diverse locations in terms of soil characteristics, climate, etc. over multiple years to develop a more effective model for N recommendation to optimize N use efficiency.