Active crop canopy sensors improve nitrogen use efficiency in dryland maize

Active canopy crop sensor commercialization offers producers the ability to vary nitrogen applications in real time based on crop reflectance measures. However, adoption of active canopy crop sensors has been limited due to inconsistent results, potential yield losses, and lack of information from field-scale trials under different management strategies. Therefore, the purpose of this study was to evaluate the capability of active crop canopy sensor system (OptRx™, Ag Leader, Ames, IA) in field trials on nine non-irrigated maize (Zea mays L.) sites in eastern Nebraska, where rainfall often limits yield (2019–2020). The sensor-based N management treatments were compared to each site’s grower treatment, examining the effects of N base rate, in-season application timing, and spatial variability technology performance. The sensor-based N management reduced N application by an average of 38.7 ± 20.8 kg Nha^-1 without a yield penalty in 77% of the sites (n = 9). The base rate of N applied prior to the in-season, sensor-based application rate in-season, and timing of the in-season application influenced the N use efficiency (NUE) of the sensor-based N management approach. Partial factor productivity of N was improved by 16.8 ± 8.4 kg grain kg N−1 relative to growers’ current management. In terms of profit, 35% of sites demonstrated a profit advantage in sensor-based treatments. Field-scale research demonstrates that active canopy sensors can improve nitrogen management efficiency and profitability. These findings highlight the importance of evaluating active canopy crop sensors under variable field conditions to optimize sensor-based N management strategies.

1 Introduction

While N fertilizer is essential for maintaining high crop production (Mosier et al., 2004), excessive applications can result in negative environmental impact and reduced profits (Schepers et al., 1997; Scharf et al., 2011; Zhang et al., 2015). Determining the optimal N rate is challenging due to the temporal and spatial variation in crop production, N loss, and N mineralization along with the dynamic interactions between soil and water (Mamo et al., 2003; Scharf et al., 2005; Shanahan et al., 2008). This has led to significant interest in site-specific nutrient management to address this spatial and temporal variability (Blackmer and White, 1998; Scharf et al., 2002; Muschietti-Piana et al., 2018; Clark et al., 2020).

With these challenges, the use of technology has the potential to improve nitrogen use efficiency (NUE) and profitability for producers in corn production (Kent Shannon et al., 2018). A variety of prediction methods have been developed across the Midwest to estimate the optimal N rate for a site using management zones and high-resolution data layers in combination with software programs to improve the accuracy of site-specific management (Fleming et al., 2000; Roberts et al., 2012). These methods include yield prediction models (Sibley et al., 2014), maximum return to N calculator (Sawyer et al., 2006), soil sampling (Sawyer and Mallarino, 2017; Morris et al., 2018; Ransom et al., 2020), and N models such as Maize-N (Setiyono et al., 2011) and Adapt-N (Sela et al., 2016). Specifically, Ransom et al. (2020) found that across 31 different corn N recommendation strategies in the Midwest, none of these tools were reliable across the entire region over many years. Thus, highlighting the challenges presented by these methods includes the amount of data required for an accurate recommendation, lack of temporal variability adjustments, averaging of spatial variability, or lack of accuracy for a range of environmental conditions. These challenges are particularly important in rainfed environments, where limited water availability makes the optimal timing and rate of N applications highly dependent on the year’s rainfall amount and distribution (Abebe and Feyisa, 2017). Thus, emphasize the importance of prediction models to account for variable environmental conditions, as well as a method to integrate this information with a sensor-based system for improved performance (Clark et al., 2020; Bean et al., 2018a, b; Thompson et al., 2015).

Active crop canopy sensors have been extensively researched since the 1990s for their ability to use real-time reflectance data to guide site-specific N management under variable environmental and seasonal conditions (Blackmer and Schepers, 1995; Bausch and Duke, 1996; Dellinger et al., 2008; Schmidt et al., 2011; Colaço and Bramley, 2018). To account for the spatial variability within a site, these systems use vegetation indices as indicators for N demand for on-the-go variable rate N applications (Raun et al., 2005; Scharf et al., 2002; Shanahan et al., 2008). The GreenSeeker™ Green 506 and Crop Circle™ ACS-210 (Holland Scientific, Lincoln, NE) have been field tested and reviewed with promising results to improve NUE (Barker and Sawyer, 2010; Colaço and Bramley, 2018). Research has focused on not only sensor types but also on refining algorithms behind them, including improving target N rate estimates (Franzen et al., 2016), optimizing application timing (Samborski et al., 2009), and developing response models such as the Holland and Schepers (2010) sufficiency index-based approach. It calculates a sufficiency index (SI) by dividing the vegetation index of the target plants by that of a well-fertilized reference while considering any N credits such as irrigation water nitrates, legume credits, previously applied N, and manure applications. While research algorithm development and calibration are ongoing (Barker and Sawyer, 2010; Whelan et al., 2012; Colaço and Bramley, 2018), accurate field-specific calibration remains essential, especially in rainfed environments where non-N factors such as soil moisture, hybrid differences, disease presence, and environmental conditions are likely to influence accurate SI value. Recent work has incorporated real-time soil and weather data, substantially improving sensor recommendations and alignment with the actual economically optimal nitrogen rate (EONR) (Colaço and Bramley, 2018; Bean et al., 2018b). Thus, reinforcing the need for environmentally responsive algorithms to enhance sensor-based management in water-limited conditions.

Given the variability in field conditions and crop responses, on-farm research has been critical for assessing the accuracy, adaptability, and potential for broader adoption of a sensor-based approach for N management. Early on, Scharf et al. (2011) developed this methodology with a large field-scale study on 55 on-farm research sites where sensor-based N strategies were compared to growers’ current N strategies, resulting in an increase in partial profit of $42 ha⁻¹ from both an increase in yield and decrease in N applied in comparison to the producer. Kitchen et al. (2010) elaborated on this concept to evaluate active crop canopy reflectance-based N application compared to growers’ current practices to evaluate profitability, resulting in $25–$50 per hectare, depending on N fertilizer cost, corn price, and soil type. However, producers are more likely to adopt technology that increases their yield ceiling than technology that may lower their input costs, as most N technologies do (Zhang et al., 2015). Therefore, the improvements of NUE and reduction of environmental impacts of active crop canopy sensors have been well-documented; much less is known about the consistencies in economic returns to the producer (Colaço and Bramley, 2018). Especially in water-limited sites, where the remote sensing techniques raise concerns that water stress may confound nitrogen stress readings (Barnes et al., 2000). In early growth stages, sites with N stress were correlated to many vegetation indices, but in rainfed sites where water is limited, there was little correlation to N stress. Previous studies have noted that these systems need to be tested for separating water stress effects on non-irrigated sites, on-farm applied research instead of simulated results, and understanding spatial variability for evaluating results at a field-length level (Colaço and Bramley, 2018; Samborski et al., 2009; Hatfield et al., 2008).

Therefore, the primary objective of this study was to evaluate the ability of active crop canopy sensors to improve NUE and profitability compared to growers’ current N management practices in non-irrigated corn fields in a humid continental climate. Specifically, the OptRx sensor system (Ag Leader Technology, Ames, IA), based on a modified Holland-Schepers model (Holland and Schepers, 2010), was used to calculate the recommended N target rates from a vegetative index. This approach was applied to examine the impact of in-season application timing, the N base rate, and the difference between the user-estimated optimum N rate and the end-of-season NUE in an active sensor-based system at non-irrigated sites across Nebraska. The secondary objective of this study was to quantify how soil spatial variability, sufficiency index, and rainfall influenced sensor-based management performance. By clarifying the functional role of active canopy crop sensors, this research aims to advance sensor-based approaches to on-farm nitrogen management.

2 Methods

2.1 Research fields

Nine site years (2019–2020) of dryland corn on-farm research experiments were located in eastern Nebraska, USA. Sensor-based N management was evaluated at five sites in 2019 (Sites 1–5) and at four sites in 2020 (Sites 6–9, Figure 1A). Each site was predominantly silt loam and silty clay loam soil types (Table 1). Annual rainfall ranges from 283.5 to 745.7 ml across the sites. The average temperature ranged from −3.31 °C to 31.66 °C throughout the growing season.

Figure 1

Figure 1. Overview of experimental sites and design. (A) A map of Nebraska showing all study sites. (B) Example of field layout illustrating the experimental design, including nitrogen (N) rate ramp treatments and N reference plots.

Table 1

Table 1. Site ID, county, year, coordinates, soil type and producer management for all nine experimental sites in eastern Nebraska (USA).

2.2 Experimental treatments

The factor evaluated was N management strategy with two levels: grower treatment and active sensor-based N treatment. All treatments were replicated six times except for Sites 6, 9, and 1, which contained five replications. The replications were arranged in a randomized complete block design. All prior field management decisions, such as tillage practices, the dates of field operations, hybrid selection, and other management practices, were made by the field owners (Table 1).

The growers’ treatment was applied by the site’s cooperating producer, and details are presented in Table 2. A base rate of N was applied prior to or at planting to all experimental sites with rates ranging from 33.6 to 156.8 kg N ha⁻¹ to establish nitrogen rate blocks (Figure 1B). This allowed for the calculation of EONR after harvest as a metric to benchmark treatment performance. The active sensor treatment consisted of split N applications directed once between V8 and V14 with the appropriate sensor system parameters (see Supplementary Table S1). In the active sensor treatment, a base rate of N was applied at least two weeks prior to the sensor-based N application. This base rate of N ranged between 39.2 and 84.0 kg N ha⁻¹ depending on the grower’s particular N management program. Base rates of each site were grouped into two categories of “Low,” representing base rates between 39 and 45 kg ha⁻¹ and the “High” classification, representing rates between 78 and 121 kg ha⁻¹.

Table 2

Table 2. Nutrient management practices include applications, dates, chemical sources, and growth stage at time of application.

Each active sensor treatment N application occurred between the V8 and V12 corn growth stages (Table 3) and was applied using a high-clearance N applicator (DTS-10, Hagie Manufacturing Company, Clarion, IA, USA) with drop tubes. The rate controller consisted of a commercially available system (PinPoint, Capstan Ag, Topeka, KS), with pulse-width modulation (PWM) nozzle solenoid valves to adjust to the changes in target N rate. In 2019, most of the applications occurred near the V12 growth stage, and in 2020, the majority of the applications occurred near the V9 growth stage. The shift to apply earlier in 2020 was made to increase the probability of the site receiving rainfall to incorporate the N application; there is a greater frequency of precipitation approximately one week post application (see Supplementary Table S2), which corresponds to the V9 growth stage (Shulski, 2020). Across all sites, liquid urea ammonium nitrate (UAN) was applied with the high-clearance N applicator for the sensor-based treatments. In 2020, a N pronitridine stabilizer, (Nitrain Bullet™, Loveland Products, Inc., Loveland, CO), was incorporated into the UAN to reduce potential losses to N volatilization.

Table 3

Table 3. Site results summary table with comparison between growers’ and the active sensor performance for yield, Preplant N, in season Applied N, total N, and partial profit for each field site in 2019 and 2020.

In this study, active crop canopy sensors (OptRx^®, AgLeader, Ames, IA) were used to calculate the vegetative index, Normalized Difference Red Edge (NDRE) (Equation 1):

$\begin{array}{ll}NDRE= \frac{NIR-RE}{NIR+RE}& (1)\end{array}$

where

$NDRE=NormalizedDifferenceRedEdge$

$NIR=NEAR–InfraredWavelength$

$RE=RedWedgeWavelength$

and to compute a recommended N rate (Equation 2):

$\begin{array}{ll}{N}_{APP}=(N_{OPT}-N_{PreFert}-N_{CRD}) \times \frac{\sqrt{(1-SI)}}{\Delta SI}& (2)\end{array}$

where

$$N_{APP}=Napplicationrate$$

$$N_{OPT}=the EONR or the maximum N rate prescribed by producers$$

$$N_{PreFert }=the sum of fertilizer N applied before sensor-based N application.$$

$$N_{CRD}=N credit for previous crop, N{O}_{3-}in irrigation water, manure application, etc.$$

$$SI=Sufficiency Index for target crop.$$

$$\begin{array}{c}\Delta SI=1-S(0);the difference betwteen SI\hfill \\ =1 and the y-intercept of the N response curve;\\ set to default of 0.7\end{array}$$

The control monitor (Integra, AgLeader, Ames, IA) records NDRE, target N rate, and applied rate in an applied file that was used for treatment implementation control and data analysis.

The OptRx sensor setup requires specific user inputs prior to application. These inputs included corn growth stage, hybrid, estimated N optimum (Nopt), N previously applied, N credits, and the minimum and maximum allowable N rate (Equation 2). The Nopt for each field was determined using a simplified University of Nebraska–Lincoln N algorithm for corn grain without accounting for soil nitrates (Shapiro et al., 2019, Supplementary Table S1). Other Nopt estimation methods were explored on the sites from 2019, including using Maize-N (Setiyono et al., 2011), a simplified UNL algorithm with grower yield goal, the full UNL algorithm (Shapiro et al., 2019), and the simplified UNL algorithm with Hybrid Maize (Yang et al., 2006) to estimate yield goal (Supplementary Table S1).

The yield goal used in these Nopt estimation methods was provided by the producer or calculated from average historical yield data and multiplied by a factor of 1.05 (Dobermann and Shapiro, 2004). The N credits in the sensor system parameters should include any N expected to be applied prior to the sensor application as well as any residual N from previous crops (i.e., soybean). For all sites, that parameter was set at zero to maintain methodological consistency and comparability among sites. As requested by the cooperating producers, the monitor settings for minimum N rate were set at 33.6 kg N ha⁻¹ and the maximum N rate was set at 336 kg N ha⁻¹. A sufficiency N strip was placed as a reference for the active sensor treatment. The high-N reference was established at least two weeks prior to the active sensor treatment application to ensure incorporation and N sufficiency, and the NDRE of this area is referred to as the reference NDRE (refNDRE) (Holland and Schepers, 2013). This high-N reference is used to create a sufficiency index (SI) (Equation 3):

$\begin{array}{ll}Sufficiency Index (SI)= \frac{NDRE}{refNDRE}& (3)\end{array}$

where

$$0\le SI \ge 1$$

$$NDRE=NDRE of target crop$$

$$refNDRE=NDRE of high-N reference$$

The rate and timing of high-N reference strip establishment and refNDRE at the time of in-season application are included for each site (Table 2).

2.3 Field data collection

2.3.1 Soil data

The coefficient of variation (CoV) in site characteristics, soil electoral conductivity and site elevation were used to assess their influence on site performance between the active sensor-based treatments and the growers’ treatments. To do so, soil electrical conductivity data was collected for each site prior to planting using an electromagnetic sensor (DUALEM-21S, Milton, ON, Canada) at 1 m and 2 m depths. Elevation data were collected from the United States Geographical Survey LIDAR dataset at a 1/3 arc-second digital elevation model (DEM) resolution. The slope of each field site’ treatment area was calculated from the DEM using the Slope toolbox in ArcGIS (ArcMap v10.6.1, ESRI, Redlands, CA, USA). Site variability in soil electrical conductivity (EC) and elevation was characterized by using coefficient of variance (CoV) (Coefficient of Variation, 2008) (Equation 4) and compared to the CoV of response variables, such as N applied and NUE, to reduce the influence of data point quantity.

$\begin{array}{ll}Coefficient of Variation (CoV)= \frac{\text{σ}}{\text{μ}}& (4)\end{array}$

where

$$\sigma =population standard deviation$$

$$\mu =population mean$$

2.3.2 Yield data

Yield data were collected using calibrated yield monitors on the grower’s combines and were post-processed using Yield Editor v 2.0.7 (USDA-ARS, Columbia, MO) to adjust for flow delay, moisture delay, maximum and minimum flow velocity, minimum swath width, maximum and minimum yield, overlap at 50% at 0.3-m cell size, and a standard deviation at three standard deviations and five header widths (Sudduth and Drummond, 2007). The harvested weight was manually adjusted for 15.5% grain moisture during post-processing, following established methods (Crowther et al., 2023).

2.4 Data analysis

N as-applied data and clean yield monitor data were spatially joined and averaged within treatment polygons labeled with replications using the analysis toolbox in ArcMap software (Esri, 2019). The results from this analysis were based exclusively on fertilizer-applied nitrogen. For each polygon, the total N applied and yield were used to calculate partial factor productivity (PFP) using Equation 5 (Kalinova et al., 2014):

$\begin{array}{ll}\text{Partial} \text{Factor Productivity }(\text{PFP})=\frac{\text{kg grain}}{kgNfertilizerapplied}& (5)\end{array}$

Partial profit was calculated within treatment blocks using yield gain or loss at the price of corn minus the increase or decrease of N applied at the price of N for a particular site (Equation 6, Kitchen et al., 2022). The cost of adopting this technology, including the sensors, machinery, and application equipment, was not included in this partial profit analysis. In 2020, the prices used were $0.138 U.S.$ kg corn⁻¹, $0.904 U.S.$ kg UAN- N⁻¹, and. $0.706 U.S.$ kg anhydrous ammonia- N⁻¹. In 2019, the prices used in the EONR calculation were $0.151 U.S.$ kg corn⁻¹, $0.794 U.S.$ kg UAN- N⁻¹, and. $0.706 U.S.$ kg anhydrous ammonia- N⁻¹. The previous formula is adjusted to each grower’s site to accommodate for variations in base rate and the cost of products applied.

$\begin{array}{ll}\begin{array}{c}\text{Partial proft}=(\text{Output Quantity }(\text{kg corn})\times \text{Output Price}(\text{U}.\text{S}.{\text{\$ kg corn}}^{-1})-\hfill \\ [(\text{Semsor and or Grower N rate }({\text{kg ha}}^{-1})+\text{Base N rate }\\ ({\text{kg ha}}^{-1}))\times \text{N Fertilizer Price }(\text{U}.\text{S}.{\text{ kg N}}^{-1})]\end{array}& (6)\end{array}$

To make comparisons between the sites, the inherent yield differences between fields from other management practices or environmental factors were removed by comparing the differences between the growers’ treatment and the sensor-based treatment. All reported values for each metric evaluated are the active sensor values minus the growers’ values. Overall results comparing treatments and summarized by sites were analyzed using a linear mixed-effects model with a significance level designated at ρ = 0.05 unless otherwise stated. Statistics were computed using R (R Core Team, 2020) for running linear mixed-effects models (Bates et al., 2020; Kuznetsova et al., 2017; Lenth, 2021), plotting data (Kassambara, 2020; Wickham, 2016; Wickham et al., 2020; Hothorn et al., 2021), and processing imagery and spatial files (Bivand, 2020; Bivand and Rundel, 2020; Hijmans, 2020; Pebesma and Bivand, 2021).

Another component of the analysis was to evaluate the influence of soil spatial characteristics on the PFP and partial profit results. To better capture the spatial differences, each treatment strip was divided into smaller 30-m length strips, and each data layer, including the as-applied N data, the yield, soil EC, site elevation, and site slope, was summarized by the mean within each treatment strip. Linear regressions were run to explore the correlation of these site characteristics to the active sensor treatment results.

3 Results

3.1 Active crop canopy sensor management compared to growers’ N management

3.1.1 Effect of sensor-based management on rate of N applied, yield, and NUE

To assess the effect of sensor-based management of the rate of N applied, yield, and NUE, the difference in active sensor treatments and grower treatments was compared for applied N and PFP for each site. In all nine of the sites for 2019 and 2020, less N was applied on average with the active sensor treatment than the grower’s treatment (Figure 2A). Reduction in N application ranged from 10 to 75 kg N ha⁻¹ while on average the reduction in N applied was 38.7 ± 20.8 kg N ha⁻¹ (Figure 2A).

Figure 2

Figure 2. N applied and PFP by site. (A) The difference in N applied displayed by site and (B) the difference in PFP displayed by site (B). Difference is determined by the average active sensor treatment minus the grower treatment. Whiskers represent standard error in the replications. Statistical significance is represented by asterisks above and below the bars (ρ* = 0.1, ρ** = 0.05, and ρ*** = 0.01). Sample sizes (n) differ among sites and treatments. All data used for calculating the represented differences can be found in Table 3.

The NUE was evaluated using the PFP of the fertilizer applied as calculated in (Equation 5). Even though there were recorded losses in yield, NUE improved on eight of the nine sites with an overall average improvement of 16.8 ± 8.4 kg grain per kg N⁻¹ and a maximum improvement of 30 kg grain per kg N⁻¹ (Figure 2B).

Since yield is often a critical factor in optimizing profitability and improving NUE, active sensor treatments and grower treatments were also compared for each site. The results show that in seven of the nine sites there were no statistical differences in yield. However, at site 3 and site 8, the yield was significantly reduced by 15.9–22.6 kg ha⁻¹ (see Supplementary Figure S1, ρ < 0.05). Across all the sites, the average difference between the active sensor treatment and the grower’s treatment was −488 ± 689 kg ha⁻¹.

The yield response to N applied was plotted for each site using the means of treatment blocks and randomized static rate blocks. At each site, the average yield resulting from the average N rate applied by the active sensors was then compared to the yield resulting from the sensor rate block of a higher N rate. These two N-rate responses were then also compared for a statistical difference in profitability. In seven of the nine sites, the active sensors’ average N rate resulted in the same or greater profit than the profit derived from the next higher rate block, indicating the active sensor treatment likely did not apply enough N to reach EONR (Table 3, ρ = 0.10). In the sites where the active sensor’s applied rate of N was less than the sensor rate blocks but resulted in a greater yield than those same rate blocks, the difference can be attributed to the active sensors distributing the N rate based on N demand spatially. The proper distribution of N rates was able to increase yield without increasing the overall N applied. Similarly, at sites where the CoV of EC and elevation was greater, the CoV of N applied was also greater. Thus, demonstrating how the active canopy crop sensors responded to variation in crop biomass as reflected by the underlying factor of site and soil variability (Supplementary Figure S3, R² ≥ 0.50; ρ ≤ 0.05).

3.1.2 Effect of sensor-based management on partial profit

To further analyze the relationship between profit and PFP (kg-grain kg-N⁻¹), the difference between the active sensor treatment and the growers’ treatment for each variable was computed to measure loss within each site replicate (Figure 3). Overall, the average difference in profit across all the sites between the active sensor and grower treatment was −$2.40 ± 15.48 US $ ha⁻¹ (−$5.93 US $ ac⁻¹). Given this, one-third of the sites were in the top right quadrant, where the active sensor treatment resulted in a greater partial profit and greater NUE than the grower’s treatment. The other two-thirds of the sites are in the bottom right quadrant, where the active sensors resulted in greater NUE but a loss in partial profit compared to the grower’s treatment. Only one of these sites, SITE 8, resulted in a significant loss of profit from sensor-based treatment (Figure 3, ρ ≤ 0.05). Last, the left two quadrants display sites where the NUE is lower than the grower’s treatment, and although a few replications had this result, no sites resulted in an average loss in NUE.

Figure 3

Figure 3. Comparison of loss metrics in partial profit and partial factor productivity. Each treatment is represented by Site (squares) and replications (circles) for each site. Loss metrics are calculated as the difference between active sensor treatments and grower treatments in partial profit and partial factor productivity. Each quadrant is separated by dotted black lines and win – loss percentage parameters are placed at the top left of each quadrant. Sample sizes (n) differ among sites and treatments.

4 Discussion

The active sensor system applied an average N rate lower than the estimated optimal N rate for the SITE 8 and SITE 9 sites, which may have resulted from underestimated crop N demand (Nopt) at the time of application. Applying the optimal N rate across a field depends on more than an accurate estimate of Nopt; the SI values across the field also contribute to the resulting target N rate. Berntsen et al. (2006) and Colaço and Bramley (2018) described this concept of redistribution of N, where the entire field will average the same amount of N as a uniform flat rate. However, areas of low or high biomass production, depending on the algorithm used, receive less N, and the medium biomass production areas receive more. To explore this redistribution of the Holland-Schepers model (embedded in the OptRx system), the N target rates across each site were compared to the Nopt parameter minus credits for each site. Across all the sites, the average N rate applied was 22.67 kg N ha⁻¹ less than the Nopt minus N credits, two values entered into the OptRx™ system. Most N rates within each site were also below this threshold, suggesting that the Nopt and N credit variables used in these studies contributed to lower N rates overall. These results further support why the active sensor treatments consistently applied less N compared to the growers’ treatments and indicate that increasing the Nopt value used in the system may be warranted.

Across all sites, sensor-based treatments tended to reduce NUE at high base N sites compared to the growers’ current treatment, while sites with low base N showed little change (differences near zero; see Supplementary Figure S2). However, the distribution of NUE differences was not normal (Wilcoxon test, ρ > 0.05), indicating that high base N sites more consistently experience lower NUE under sensor-based treatments than low base sites. To assess the effect of base N rates on the performance of the active sensor system, SITE 6 included a comparison of two base rates (39.2 kg ha⁻¹ vs. 78.4 kg ha⁻¹) applied at two growth stages (V8 growth stage vs. V11 growth stage). The results show an increased distribution of total N rates between the replications at the lower N base rates. This is because the sensor-based system has a greater Nopt-N pre-value (i.e., a greater range of N for the algorithm to operate within) at a lower base rate than a higher base rate. Therefore, SI has a greater influence on the total N applied. Thus, indicating the active sensor system was able to compensate for the difference in base N rates to apply nearly the same amount of total N within the same application timing. A similar result occurred in Thompson and Puntel (2020), where a UAV-based N management had two treatments with the same total N applied following two differing base rates.

Since economic optimum nitrogen rate (EONR) varies throughout a field from varied soil nitrate concentrations, soil characteristics, landscape position, soil-water interactions, and crop N demand (Ransom, 2018; Wang et al., 2020; Crowther et al., 2023). From this, fields of greater spatial variability would have greater variability in EONR values and would benefit from sensor-based variable rate technology. In this study, it was found that the sites with less variability in elevation and soil EC resulted in the greatest differences in N applied and NUE. Further demonstrating how active crop canopy sensors responded to underlying factors like site and soil variability, not just variability in crop biomass. Also suggesting that more homogenous sites (lower CoV) tended to show greater declines in NUE under the evaluated conditions, while more heterogeneous sites displayed more neutral NUE responses. While these results contradict the notion that sites with greater spatial variability would achieve higher NUE, improved NUE—when N rates are at or below EONR—still contributes to reducing potential N losses to the environment (Zhao et al., 2016; Hong et al., 2007). Further research is needed to explore this spatial variability to explain this result.

Overall, the influence from differences in timing, method, or source (i.e., products) of the application between the grower’s treatment and the active sensor treatment influenced the yield and partial profit. Particularly in one site, Site 9, where both the grower treatment yield and the yield of the grower rate blocks resulted in a lower yield than the sensor rate block of similar rates. At this site, the grower treatment evidently experienced more N losses from the pre-plant N application timing than the in-season N application with the sensor-based system. These results reflect those of Scharf et al. (2011) and Raun et al. (2005), both of whom observed that sensor-based treatments produced greater yields when their N rate exceeded that of the growers’ treatment. In consideration of PFP, three sites, Sites 2, 3, and 8, the grower N rate resulted in a statistically higher partial profit than the rate blocks established during the active sensor application of the same or greater N rate. For example, SITE 8, which experienced rainfall during the growing season, was greater than the 30-year normal precipitation for that field (496.3 > 402.2). Therefore, yields were greater than average. From these results, it can be concluded that other factors, such as N source, differences in N losses, and precipitation, also influenced the yield and partial profit in the treatment comparisons, which is consistent with Spackman et al. (2019). This provides evidence for the importance of understanding the management system surrounding a sensor-based strategy in non-irrigated fields for this technology adoption.

5 Conclusion

The aim of this study was to evaluate the ability of active crop canopy sensors to improve NUE and profitability compared to growers’ current N management practices in non-irrigated cornfields in a humid continental climate. In consideration of the impact of in-season application timing, the N base rate, and NUE across all sites, an active sensor-based system applied less N than the growers’ current management (38.7 ± 20.8 kg N ha⁻¹). The sensor-based treatment was able to vary the distribution of a similar total N rate according to the crop-specific needs. Accounting for within-field variability improved NUE without yield loss in 77% of the site years, suggesting the sensor-based approach captures spatial in a way that benefits producers. Compared to a fixed N rate applied on the same date and method, sensor-based treatments were more profitable at seven of nine sites, with two sites showing a nonsignificant loss ($2.40 ± 15.48 U.S. $ ha⁻¹). Other factors such as rainfall following application, average SI at the time of application, and soil variability did not have a direct correlation to the profitability of sensor-based technology. Therefore, these results demonstrate the application timing, source, and method all greatly influence the N response, especially in non-irrigated silt loam soils of eastern Nebraska.

While N fertilizer remains a critical crop input, determining the optimal N rate is further complicated by the spatial and temporal variability in non-irrigated maize production (Mosier et al., 2004; Scharf et al., 2011; Zhang et al., 2015). Variability that drives differences in field conditions and crop responses highlights the importance of on-farm research to critically evaluate the broad-scale adoption of sensor-based approaches for N management in maize production. Overall, as site variability increased, as characterized by the coefficient of variation of soil EC and site elevation, the variation of N applied using the sensor-based system also increased. However, the insights from this study show the active sensor systems were able to compensate for the differences in base N rates to apply nearly the same amount of total N within the same application timing. Continued efforts are needed to improve the understanding of how the conditions under which sensor-based N management is most effective to support broader on-farm adoption.

Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Author contributions

SK: Investigation, Formal Analysis, Project administration, Data curation, Visualization, Writing – original draft, Conceptualization, Writing – review & editing, Methodology. KB: Visualization, Writing – review & editing. JL: Supervision, Writing – review & editing, Resources, Methodology, Validation, Funding acquisition, Project administration. LP: Writing – review & editing, Investigation, Methodology, Supervision. LT: Investigation, Supervision, Writing – review & editing, Methodology. GB: Writing – review & editing.

Funding

The author(s) declared financial support was received for this work and/or its publication. Funding for this research project was provided by the Nebraska Corn Board. We would also like to acknowledge support for this project from USDA-ARS under project agreement # 58-3042-1-014. USDA is an equal opportunity provider and employer. Mention of trade names or commercial products in this publication does not imply recommendation or endorsement by the U.S. Department of Agriculture.

Acknowledgments

We thank the cooperating producers for their support and collaboration in providing access to their field sites, which was essential to this research. We also thank Jackson Stansell and Tyler Smith for their support in data collection, field management, and overall project support.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declare that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fagro.2025.1722488/full#supplementary-material

References

Abebe Z. and Feyisa H. (2017). Effects of nitrogen rates and time of application on yield of maize: rainfall variability influenced time of N application. Int. J. Agron. 2017, 1545280. doi: 10.1155/2017/1545280

Crossref Full Text | Google Scholar

Barker D. W. and Sawyer J. E. (2010). Using active canopy sensors to quantify corn nitrogen stress and nitrogen application rate. Agron. J. 102, 964–971. doi: 10.2134/agronj2010.0004

Crossref Full Text | Google Scholar

Barnes E., Clarke T., Richards S., Colaizzi P., Haberland J., Kostrzewski M., et al. (2000). Coincident detection of crop water stress, nitrogen status and canopy density using ground based multispectral data. 1619, 6. In Proceedings of the fifth international conference on precision agriculture. (Bloomington, MN, USA) 1619, 6

Google Scholar

Bates D., Maechler M., Bolker B., and Walker S. (2020). lme4: Linear mixed-effects models using Eigen 452 and S4 (GitHub). Available online at: https://github.com/lme4/lme4/ (Accessed February 10, 2021).

Google Scholar

Bausch W. C. and Duke H. (1996). Remote sensing of plant nitrogen status in corn. Transactions ASAE 39, 1869–1875.

Google Scholar

Bean G. M., Kitchen N. R., Camberato J. J., Ferguson R. B., Fernandez F. G., Franzen D. W., et al. (2018a). Active-optical reflectance sensing corn algorithms evaluated over the United States midwest corn belt. Agron. J. 110, 2552–2565. doi: 10.2134/agronj2018.03.0217

Crossref Full Text | Google Scholar

Bean G. M., Kitchen N. R., Camberato J. J., Ferguson R. B., Fernandez F. G., Franzen D. W., et al. (2018b). Improving an active-optical reflectance sensor algorithm using soil and weather information. Agron. J. 110, 2541–2551. doi: 10.2134/agronj2017.12.0733

Crossref Full Text | Google Scholar

Berntsen J., Thomsen A., Schelde K., Hansen O. M., Knudsen L., Broge N., et al. (2006). Algorithms for sensor-based redistribution of nitrogen fertilizer in winter wheat. Precis. Agric. 7, 65–83. doi: 10.1007/s11119-006-9000-2

Crossref Full Text | Google Scholar

Bivand R. (2020). spdep: Spatial dependence: weighting schemes, statistics (GitHub) pp. 464. Available online at: https://github.com/r-spatial/spdep/ (Accessed February 10, 2021).

Google Scholar

Bivand R. and Rundel C. (2020). rgeos: Interface to geometry engine – open source (GEOS) (R-Forge). Available online at: https://r-forge.r-project.org/projects/rgeos/https://trac.osgeo.org/geos/http://rgeos.r-forge.r-project.org/index.html.

Google Scholar

Blackmer T. and Schepers J. (1995). Use of a chlorophyll meter to monitor nitrogen status and schedule fertigation for corn. J Production Agricul. 8, 56–60.

Google Scholar

Blackmer A. M. and White S. (1998). Using precision farming technologies to improve management of soil and fertiliser nitrogen. Aust. J. Agric. Res. 49, 555–564. doi: 10.1071/A97073

Crossref Full Text | Google Scholar

Clark J. D., Fernández F. G., Camberato J. J., Carter P. R., Ferguson R. B., Franzen D. W., et al. (2020). Weather and soil in the US Midwest influence the effectiveness of single- and split-nitrogen applications in corn production. Agron. J. 112, 5288–5299. doi: 10.1002/agj2.20446

Crossref Full Text | Google Scholar

Coefficient of Variation (2008). The Concise Encyclopedia of Statistics (New York, NY: Springer New York), 95–96. doi: 10.1007/978-0-387-32833-1_65

Crossref Full Text | Google Scholar

Colaço A. F. and Bramley R. G. V. (2018). Do crop sensors promote improved nitrogen management in grain crops? Field Crops Res. 218, 126–140. doi: 10.1016/j.fcr.2018.01.007

Crossref Full Text | Google Scholar

Crowther J., Parrish J., Luck J. D., and Ferguson R. B. (2023). Evaluating management zones and crop-sensing relationships for improved irrigated maize nitrogen management. Agrosystems Geosciences Environ. 6, e20336. doi: 10.1002/agg2.20336

Crossref Full Text | Google Scholar

Dellinger A. E., Schmidt J. P., and Beegle D. B. (2008). Developing nitrogen fertilizer recommendations for corn using an active sensor. Agron. J. 100, 1546–1552. doi: 10.2134/agronj2007.0386

Crossref Full Text | Google Scholar

Dobermann A. and Shapiro C. A. (2004). Setting a realistic corn yield goal. (NE, United States: Cooperative Extension, Institute of Agriculture and Natural Resources).

Google Scholar

Esri. (2019). Arcmap: Release 10.9. (Redlands, CA, United States: Environmental Systems Research Institute). Available online at: https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview. (Accessed February 3, 2021).

Google Scholar

Fleming K. L., Westfall D. G., Wiens D. W., and Brodahl M. C. (2000). Evaluating farmer defined management zone maps for variable rate fertilizer application. Precis. Agric. 2, 201–215. doi: 10.1023/A:1011481832064

Crossref Full Text | Google Scholar

Franzen D., Kitchen N., Holland K., Schepers J., and Raun W. (2016). Algorithms for in-season nutrient management in cereals. Agron. J. 108, 1775–1781. doi: 10.2134/agronj2016.01.0041

Crossref Full Text | Google Scholar

Hatfield J. L., Gitelson A. A., Schepers J. S., and Walthall C. L. (2008). Application of spectral remote sensing for agronomic decisions. Agron. J. 100, S–117. doi: 10.2134/agronj2006.0370c

Crossref Full Text | Google Scholar

Hijmans R. J. (2020). raster: Geographic data analysis and modeling (R Spatial). Available online at: https://rspatial.org/raster (Accessed February 3, 2021).

Google Scholar

Holland K. H. and Schepers J. S. (2010). Derivation of a variable rate nitrogen application model for in-season fertilization of corn. Agron. J. 102, 1415–1424. doi: 10.2134/agronj2010.0015

Crossref Full Text | Google Scholar

Holland K. H. and Schepers J. S. (2013). Use of a virtual-reference concept to interpret active crop canopy sensor data. Precis. Agric. 14, 71–85. doi: 10.1007/s11119-012-9301-6

Crossref Full Text | Google Scholar

Hong N., Scharf P. C., Davis J. G., Kitchen N. R., and Sudduth K. A. (2007). Economically optimal nitrogen rate reduces soil residual nitrate. J. Environ. Qual. 36, 354–362. doi: 10.2134/jeq2006.0173

PubMed Abstract | Crossref Full Text | Google Scholar

Hothorn T., Bretz F., and Westfall P. (2021). multcomp: Simultaneous inference in general parametric models (CRAN). Available online at: https://CRAN.R-project.org/package=multcomp (Accessed February 3, 2021).

Google Scholar

Kalinova S., Kostadinova S., and Hristoskov A. (2014). Nitrogen use efficiency and maize yield response to nitrogen rate and foliar fertilizing. Bulgarian J. Agricul. Sci. 20, 178–181.

Google Scholar

Kassambara A. (2020). ggpubr: ggplot2 based publication ready plots (Datanovia). Available online at: https://rpkgs.datanovia.com/ggpubr/ (Accessed February 3, 2021).

Google Scholar

Kent Shannon D., Clay D. E., and Sudduth K. A. (2018). An introduction to precision agriculture,” in. Precis. Agric. Basics 1–12, 1–1. doi: 10.2134/precisionagbasics.2016.0084

Crossref Full Text | Google Scholar

Kitchen N. R., Sudduth K. A., Drummond S. T., Scharf P. C., Palm H. L., Roberts D. F., et al. (2010). Ground-based canopy reflectance sensing for variable-rate nitrogen corn fertilization. Agron. J. 102, 71–84. doi: 10.2134/agronj2009.0114

Crossref Full Text | Google Scholar

Kitchen N. R., Ransom C. J., Schepers J. S., Hatfield J. L., Massey R., and Drummond S. T. (2022). A new perspective when examining maize fertilizer nitrogen use efficiency, incrementally. Plos one 17, e0267215.

PubMed Abstract | Google Scholar

Kuznetsova A., Brockhoff P. B., and Christensen R. H. B. (2017). lmerTest package: Tests in linear mixed effects models. J. Stat. Software 82, 1–26. doi: 10.18637/jss.v082.i13

Crossref Full Text | Google Scholar

Lenth R. V. (2021). emmeans: Estimated marginal means, aka least-squares means (GitHub). Available online at: https://github.com/rvlenth/emmeans (Accessed February 3, 2021).

Google Scholar

Mamo M., Malzer G. L., Mulla D. J., Huggins D. R., and Strock J. (2003). Spatial and temporal variation in economically optimum nitrogen rate for corn. Agron. J. 95, 958–964. doi: 10.2134/agronj2003.9580

Crossref Full Text | Google Scholar

Morris T. F., Murrell T. S., Beegle D. B., Camberato J. J., Ferguson R. B., Grove J., et al. (2018). Strengths and limitations of nitrogen rate recommendations for corn and opportunities for improvement. Agron. J. 110, 1–37. doi: 10.2134/agronj2017.02.0112

Crossref Full Text | Google Scholar

Mosier A. R., Syers J. K., and Freney J. R. (2004). Nitrogen fertilizer: an essential component of increased food, feed, and fiber production. Agric. nitrogen Cycle 65, 3–15.

Google Scholar

Muschietti-Piana M., del P., Cipriotti P. A., Urricariet S., Peralta N. R., and Niborski M. (2018). Using site-specific nitrogen management in rainfed corn to reduce the risk of nitrate leaching. Agric. Water Manage. 199, 61–70. doi: 10.1016/j.agwat.2017.12.002

Crossref Full Text | Google Scholar

Pebesma E. and Bivand R. (2021). sp: Classes and methods for spatial data (CRAN). Available online at: https://CRAN.R-project.org/package=sp (Accessed February 3, 2021).

Google Scholar

Ransom C. J. (2018). Evaluating and improving corn nitrogen fertilizer recommendation tools across the US Midwest. Dissertation. (University of Missouri-Columbia, MO, United States)

Google Scholar

Ransom C. J., Kitchen N. R., Camberato J. J., Carter P. R., Ferguson R. B., Fernández F. G., et al. (2020). Corn nitrogen rate recommendation tools’ performance across eight US midwest corn belt states. Agron. J. 112, 470–492. doi: 10.1002/agj2.20035

Crossref Full Text | Google Scholar

Raun W. R., Solie J. B., Stone M. L., Martin K. L., Freeman K. W., Mullen R. W., et al. (2005). Optical sensor-based algorithm for crop nitrogen fertilization. Commun. Soil Sci. Plant Anal. 36, 2759–2781. doi: 10.1080/00103620500303988

Crossref Full Text | Google Scholar

R Core Team (2020). R: A language and environment for statistical computing (Vienna, Austria: R Foundation for Statistical Computing). Available online at: https://www.R-project.org/.

Google Scholar

Roberts D. F., Ferguson R. B., Kitchen N. R., Adamchuk V. I., and Shanahan J. F. (2012). Relationships between soil-based management zones and canopy sensing for corn nitrogen management. Agron. J. 104, 119–129. doi: 10.2134/agronj2011.0044

Crossref Full Text | Google Scholar

Samborski S. M., Tremblay N., and Fallon E. (2009). Strategies to make use of plant sensors-based diagnostic information for nitrogen recommendations. Agron. J. 101, 800–816. doi: 10.2134/agronj2008.0162Rx

Crossref Full Text | Google Scholar

Sawyer J. and Mallarino A. (2017). Use of the late-spring soil nitrate test in Iowa corn production. (Ames, IA, United States: Iowa State University Extension and Outreach)

Google Scholar

Sawyer J., Nafziger E., Randall G., Bundy L., Rehm G., and Joern B. (2006). Concepts and rationale for regional nitrogen rate guidelines for corn (Ames, Iowa: Iowa State University-University Extension), 28.

Google Scholar

Scharf P. C., Kitchen N. R., Sudduth K. A., Davis J. G., Hubbard V. C., and Lory J. A. (2005). Field-scale variability in optimal nitrogen fertilizer rate for corn. Agron. J. 97, 452–461. doi: 10.2134/agronj2005.0452

Crossref Full Text | Google Scholar

Scharf P. C., Schmidt J. P., Kitchen N. R., Sudduth K. A., Hong S. Y., Lory J. A., et al. (2002). Remote sensing for nitrogen management. J. Soil Water Conserv. 57, 518–524. doi: 10.1080/00224561.2002.12457487

Crossref Full Text | Google Scholar

Scharf P. C., Shannon D. K., Palm H. L., Sudduth K. A., Drummond S. T., Kitchen N. R., et al. (2011). Sensor-based nitrogen applications out-performed producer-chosen rates for corn in on-farm demonstrations. Agron. J. 103, 1683–1691. doi: 10.2134/agronj2011.0164

Crossref Full Text | Google Scholar

Schepers J., Moravek M., Bishop R., and Johnson S. (1997). Impact of nitrogen and water management on ground water quality. Ground water: Prot. Alternatives Strategies U.S.A. 267–268.

Google Scholar

Schmidt J., Beegle D., Zhu Q., and Sripada R. (2011). Improving in-season nitrogen recommendations for maize using an active sensor. Field Crops Res. 120, 94–101. doi: 10.1016/j.fcr.2010.09.005

Crossref Full Text | Google Scholar

Sela S., van Es H. M., Moebius-Clune B. N., Marjerison R., Melkonian J., Moebius-Clune D., et al. (2016). Adapt-N outperforms grower-selected nitrogen rates in northeast and midwestern United States strip trials. Agron. J. 108, 1726–1734. doi: 10.2134/agronj2015.0606

Crossref Full Text | Google Scholar

Setiyono T. D., Yang H., Walters D. T., Dobermann A., Ferguson R. B., Roberts D. F., et al. (2011). Maize-N: A decision tool for nitrogen management in maize. Agron. J. 103, 1276–1283. doi: 10.2134/agronj2011.0053

Crossref Full Text | Google Scholar

Shanahan J. F., Kitchen N. R., Raun W. R., and Schepers J. S. (2008). Responsive in-season nitrogen management for cereals. Comput. Electron. Agric. 61, 51–62. doi: 10.1016/j.compag.2007.06.006

Crossref Full Text | Google Scholar

Shapiro C. A., Ferguson R., Wortmann C. S., Maharjan B., and Krienke B. T. (2019). Nutrient management suggestions for corn. (Lincoln, NE, United States: University of Nebraska-Lincoln Extension Circular EC117).

Google Scholar

Shulski M. (2020). NC3 Nebraska climate summary. (Lincoln, Lincoln, NE United States: University of Nebraska).

Google Scholar

Sibley A. M., Grassini P., Thomas N. E., Cassman K. G., and Lobell D. B. (2014). Testing remote sensing approaches for assessing yield variability among maize fields. Agron. J. 106, 24–32. doi: 10.2134/agronj2013.0314

Crossref Full Text | Google Scholar

Spackman J. A., Fernandez F. G., Coulter J. A., Kaiser D. E., and Paiao G. (2019). Soil texture and precipitation influence optimal time of nitrogen fertilization for corn. Agron. J. 111, 2018–2030. doi: 10.2134/agronj2018.09.0605

Crossref Full Text | Google Scholar

Sudduth K. A. and Drummond S. T. (2007). Yield editor: software for removing errors from crop yield maps. Agron. J. 99, 1471–1482. doi: 10.2134/agronj2006.0326

Crossref Full Text | Google Scholar

Thompson L. J., Ferguson R. B., Kitchen N., Frazen D. W., Mamo M., Yang H., et al. (2015). Model and sensor-based recommendation approaches for in-season nitrogen management in corn. Agron. J. 107, 2020–2030. doi: 10.2134/agronj15.0116

Crossref Full Text | Google Scholar

Thompson L. J. and Puntel L. A. (2020). Transforming unmanned aerial vehicle (UAV) and multispectral sensor into a practical decision support system for precision nitrogen management in corn. Remote Sens. 12. doi: 10.3390/rs12101597

Crossref Full Text | Google Scholar

Wang X., Miao Y., Dong R., Chen Z., Kusnierek K., Mi G., et al. (2020). Economic optimal nitrogen rate variability of maize in response to soil and weather conditions: implications for site-specific nitrogen management. Agronomy 10. doi: 10.3390/agronomy10091237

Crossref Full Text | Google Scholar

Whelan B. M., Taylor J. A., and McBratney A. B. (2012). A ‘small strip’ approach to empirically determining management class yield response functions and calculating the potential financial ‘net wastage’ associated with whole-field uniform-rate fertiliser application. Field Crops Res. 139, 47–56. doi: 10.1016/j.fcr.2012.10.012

Crossref Full Text | Google Scholar

Wickham H. (2016). ggplot2: Elegant graphics for data analysis (New York: Springer-Verlag). Available online at: https://ggplot2.tidyverse.org (Accessed February 10, 2021).

Google Scholar

Wickham H., Chang W., Henry L., Pedersen T. L., Takahashi K., Wilke C., et al. (2020). ggplot2: Create elegant data visualisations using the grammar of graphics (CRAN). Available online at: https://CRAN.R-project.org/package=ggplot2 (Accessed February 3, 2021).

Google Scholar

Yang H., Dobermann A., Cassman K. G., and Walters D. T. (2006). Features, applications, and limitations of the hybrid-maize simulation model. Agron. J. 98, 737–748. doi: 10.2134/agronj2005.0162

Crossref Full Text | Google Scholar

Zhang X., Mauzerall D. L., Davidson E. A., Kanter D. R., and Cai R. (2015). The economic and environmental consequences of implementing nitrogen-efficient technologies and management practices in agriculture. J. Environ. Qual. 44, 312–324. doi: 10.2134/jeq2014.03.0129

PubMed Abstract | Crossref Full Text | Google Scholar

Zhao X., Christianson L. E., Harmel D., and Pittelkow C. M. (2016). Assessment of drainage nitrogen losses on a yield-scaled basis. Field Crops Res. 199, 156–166. doi: 10.1016/j.fcr.2016.07.015

Crossref Full Text | Google Scholar