A Case Study of Environmental Benefits of Sensor ...

2 downloads 93278 Views 895KB Size Report
Jan 29, 2016 - formly leads to mismatches between N fertilizer application rate and crop N need ..... country or global scale and build on simplifying assumptions, such as constant ..... Sawyer, P.C. Scharf, J.S. Strock, and M.B. Villamil. 2014.
Published January 29, 2016

Journal of Environmental Quality

TECHNICAL REPORTS Environmental Models, Modules, and Datasets

A Case Study of Environmental Benefits of Sensor-Based Nitrogen Application in Corn Ao Li, Benjamin D. Duval, Robert Anex,* Peter Scharf, Jenette M. Ashtekar, Phillip R. Owens, and Charles Ellis Abstract Crop canopy reflectance sensors make it possible to estimate crop N demand and apply appropriate N fertilizer rates at different locations in a field, reducing fertilizer input and associated environmental impacts while maintaining crop yield. Environmental benefits, however, have not been quantified previously. The objective of this study was to estimate the environmental impact of sensor-based N fertilization of corn using model-based environmental Life Cycle Assessment. Nitrogen rate and corn grain yield were measured during a sensor-based, variable N-rate experiment in Lincoln County, MO. Spatially explicit soil properties were derived using a predictive modeling technique based on in-field soil sampling. Soil N2O emissions, volatilized NH3 loss, and soil NO3- leaching were predicted at 60 discrete field locations using the DeNitrificationDeComposition (DNDC) model. Life cycle cumulative energy consumption, global warming potential (GWP), acidification potential, and eutrophication potential were estimated using model predictions, experimental data, and life cycle data. In this experiment, variable-rate N management reduced total N fertilizer use by 11% without decreasing grain yield. Precision application of N is predicted to have reduced soil N2O emissions by 10%, volatilized NH3 loss by 23%, and NO3- leaching by 16%, which in turn reduced life cycle nonrenewable energy consumption, GWP, acidification potential, and eutrophication potential by 7, 10, 22, and 16%, respectively. Although mean N losses were reduced, the variations in N losses were increased compared with conventional, uniform N application. Crop canopy sensor-based, variable-rate N fertilization was predicted to increase corn grain N use efficiency while simultaneously reducing total life-cycle energy use, GWP, acidification, and eutrophication.

Core Ideas • Efficient use and application of N fertilizer likely reduces environmentally harmful N losses. • Sensor-based N fertilization has the promise of maximizing yield while minimizing N loss. • Sensor-based fertilization maintained corn yield and reduced losses of NO3- and N2O. • Sensor-based fertilization yielded life cycle GWP, acidification, and eutrophication benefits.

Copyright © American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America. 5585 Guilford Rd., Madison, WI 53711 USA. All rights reserved. J. Environ. Qual. doi:10.2134/jeq2015.07.0404 Freely available online through the author-supported open-access option. Received 8 Aug. 2015. Accepted 11 Nov. 2015. *Corresponding author ([email protected]).

P

recision agriculture techniques, such as the use

of global positioning systems, field mapping with geographic information systems, and variable-rate application of pesticides and fertilizer, can improve crop yield and the economic efficiency of crop production (Nash et al., 2009). Precision agriculture can also yield significant environmental benefit through reduced or targeted application of nutrients and pesticides and reduction of soil compaction and erosion through control of equipment travel patterns (McLoud et al., 2007) Improved efficiencies in corn (Zea mays L.) production are of particular interest because corn is by most measures the most important crop produced in the United States. More tons of field corn are produced each year in the United States than any other crop, and field corn has more economic value than any other US crop (FAOSTAT, 2014). Corn is grown on over 400,000 US farms and accounts for almost one quarter of the total harvested crop acres in the country (USDA–NASS, 2014). Corn accounts for by far the largest fraction (37–51%) of total fertilizer consumed annually in the United States (Snyder, 2012). Although the nitrogen (N) use efficiency of corn has improved significantly in recent decades (Cassman et al., 2002), a substantial amount of applied N is still lost from the system, resulting in a variety of negative consequences. Corn agriculture has a relatively large environmental footprint due in part to the high N fertilizer requirement of the crop (Vanotti and Bundy, 1994). Conventional production of corn involves fertilizer application at or before planting many weeks before there is significant crop N uptake. This leaves fertilizer N in the soil vulnerable to loss through nitrate (NO3-) leaching or to loss to the atmosphere as volatilized ammonia (NH3) or as nitrous oxide (N2O) generated via microbial nitrification–denitrification pathways (Schlesinger, 1997). Fertilizer N loss from corn production is thus implicated in many of the most severe environmental impacts of agriculture: groundwater and surface water quality impacts through nitrate loss and eutrophication of terrestrial and coastal systems; local air quality impacts through loss as ammonia, NO, and NO2; climate change through release

A. Li, B.D. Duval, and R. Anex, Dep. of Biological Systems Engineering, Univ. of Wisconsin, Madison, WI 53716; P. Scharf, Division of Plant Sciences, Univ. of Missouri, Columbia, MO 65211; J.M. Ashtekar and P.R Owens, Agronomy Dep., Purdue Univ., West Lafayette, IN 47907; C. Ellis, Univ. of Missouri Extension, Troy, MO 63379. Assigned to Associate Editor Søren Petersen. Abbreviations: DNDC, DeNitrification-DeComposition; GHG, greenhouse gas; GWP, global warming potential; LCA, Life Cycle Assessment; LCI, Life Cycle Inventory; LCIA, Life Cycle Impact Assessment; SOC, soil organic carbon.

of N2O; and depletion of stratospheric ozone through breakdown of N2O to nitrogen oxides (Reay et al., 2012; Aneja et al., 2009; Gruber and Galloway, 2008; Ravishankara et al., 2009). Once reactive N leaves the agricultural system, its consequences grow over time as it moves along the biogeochemical pathways in what Galloway et al. (2003) call the “nitrogen cascade.” In this way the same reactive N can cause multiple effects in the atmosphere, terrestrial ecosystems, and freshwater and marine systems and on human health. The need for N fertilizer often varies widely across an individual field (Mamo et al., 2003). As a result, fertilizing fields uniformly leads to mismatches between N fertilizer application rate and crop N need (Shanahan et al., 2008). A precision agriculture approach to addressing the disparate spatial N requirements across a field is the use of a variable-rate application guided by a crop canopy reflectance sensor system. Crop reflectance sensors such as Greenseeker (NTech Industries) and the CropCircle (Holland Scientific) generate modulated light in the visible and near infrared regions, and the detected canopy reflectance is used to generate N application recommendations in real time (Walburg et al., 1982). The advantages of this approach, compared with a uniform application of fertilizer on a field, are that (i) crops needing greater N additions receive that input and (ii) fertilizer application is reduced where crop demand is lower (Scharf et al., 2005). Multiple studies have evaluated the performance of canopy sensors for detecting crop N stress with the aim of optimizing variable-rate N application (Barker and Sawyer, 2010; Dellinger et al., 2008; Kitchen et al., 2010; Scharf and Lory, 2009; Shaver et al., 2011). Normalized difference vegetation index, a common remote sensing index, has been shown to correlate with economically optimal N application rates (Schmidt et al., 2009). In a study comparing producer-determined N application rates and sensor-based N application, Scharf et al. (2011) demonstrated a simultaneous increase in corn yield and approximately 8% reduction in N fertilizer use. Lower rates of fertilizer application and greater N use efficiency are expected to lead to environmental benefits because lower residual soil N should result in lower NO3- leaching and provide less substrate for N2O production (Roberts et al., 2010). Variable-rate fertilization is recommended as a practice that can reduce greenhouse gas (GHG) emissions in several GHG emissions reduction protocols (Millar et al., 2012; Climate Action Reserve, 2013; Millar et al., 2013). Variable-rate fertilization is also included in the Global 4R Nutrient Stewardship guidelines developed by the fertilizer industry to address concerns that fertilizer use is detrimental to the environment ( Johnston and Bruulsema, 2014). By reducing total fertilizer use, variable-rate N fertilization has the potential to reduce environmental impacts associated with fertilizer production as well as those that occur after application. Variable-rate N fertilization may also create some new environmental emissions, however, because sensor-based N application occurs during vegetative growth and thus may require an additional pass through the field, resulting in additional fuel consumption and associated emissions of combustion. On a life cycle basis, variable-rate N fertilization has the potential to reduce total energy consumption and environmental impacts, but this potential has not previously been evaluated

and quantified. Here, we use a case study from Missouri to calculate the environmental performance of the sensor-based, variable-rate N fertilization of corn and compare it with a uniform producer-defined N rate using process-based modeling and life cycle analysis. To our knowledge, no previous published studies have examined the environmental impacts and life cycle trade-offs of sensor-based, variable-rate N fertilization for corn. Specifically, the goals of this study were (i) to compare N loss (soil N2O emissions, NH3 volatilization, and NO3- leaching) and yield associated with sensor-based, variable-rate N fertilization and producer-specified, fixed-rate N fertilization and (ii) to compare the life cycle environmental impacts of these fertilization systems. We assessed the following impact categories: fossil energy use, global warming potential (GWP), acidification potential, and eutrophication potential.

Materials and Methods Experimental Site A field experiment designed to compare the efficacy of sensorbased, variable-rate N fertilization of corn relative to a producerdefined N rate was established in 2008 at a cooperator farm in Lincoln County, MO (lat. 39.12, long. -91.00) (Scharf et al., 2011). The soils at the site were silt-loam Hapludalfs, and the cropping system was a no-till corn–soybean rotation. The entire field received preplant N applied as broadcast urea and diammonium phosphate at a rate of 45 kg N ha-1, chosen by the cooperating producer. Fertilizer at the producer-defined rate and variable rates was applied in parallel strips as shown in Fig. 1. A second application of N fertilizer was injected at the V7 growth stage as urea–ammonium–nitrate (Scharf et al., 2011). The producer-defined side-dress N rate was 86 kg N ha-1 (Fig. 1). A Crop Circle 210 crop reflectance sensor (Holland Scientific) was used in the application of the variable-rate sidedress fertilization based on the relative visible/near infrared reflectance ratio (i.e., visible to near infrared reflectance in the target application area divided by the visible/near infrared reflectance values in the reference area) as described by Scharf et al. (2011). Nitrogen was applied according to the sensor-based algorithm unless outside of limits requested by the cooperating farmer. The minimum application rate was 56 kg N ha-1, and the maximum application rate was 95 kg N ha-1. The N application rate and grain yield at harvest were recorded using on-board data recording and mapped using global position systems. Total N application rate varied from 101 to 140 kg N ha-1 (preplant plus side-dress application).

DeNitrification-DeComp Modeling The variation in crop N demand across a field to which a sensor-based fertilization system adapts results from variation in the soil characteristics controlling N transport and N transformation processes. It is impractical to monitor multiple environmental parameters at a large number of points in a production field, so we used the DeNitrification-DeComposition (DNDC) model to estimate N2O emission, NH3 volatilization, and soil NO3- leaching. The DNDC model is a process-oriented simulation model of carbon (C) and N biogeochemistry (Li et al., Journal of Environmental Quality

Fig. 1. Field map of experimental location in Lincoln County, MO. Strips in light blue are producer-rate fertilization strips, and strips with varied color dots are variable-rate fertilization strips. Black triangles are modeled locations of producer-defined fertilization. Pink stars are modeled locations of variable-rate N application.

1992, 1997). The DNDC model has been applied extensively to corn production systems and has accurately predicted N2O emissions and NO3- leaching (Li et al., 2006; Smith et al., 2002, 2008; Tonitto et al., 2007). For site-specific simulation as performed here, DNDC inputs include climate data, soil physical and chemical data, vegetation/crop data, and the schedule of management events. Daily weather data from 1979 to 2008 were taken from the National Weather Service Cooperative Observer Program (NOAA, 2014). Management events include preplant fertilization, planting, herbicide application, side-dress fertilization, and harvesting. The dates of most field operations were reported by the cooperating farmer. The date of harvest was estimated based on USDA-reported usual harvesting dates (USDA–NASS, 2014). Use of the DNDC model requires soil data at each simulated location in the field, but extensive soil testing throughout the field was impractical. Alternatively, we developed detailed soil data using a geomorphometric fuzzy logic soil mapping approach that creates continuous soil property predictions at the resolution of the digital elevation model (Ashtekar et al., 2014). This method used the Geomorphons add-on in Geographic Resources Analysis Support System–geographic information systems, in which all pixels on a 10-m gridded digital elevation model were classified into 1 of 10 possible landforms. The landforms were then grouped into five classes, each representing a different soil–landform association, resulting in 50 distinct soil types. Continuous property maps were then generated using a fuzzy logic approach, where ideal class criteria and class threshold values were determined statistically from the distribution of wetness index, normalized height, and % slope within each geomorphon soil–landform class. The mapping method was calibrated using measurements of soil pH, organic matter, mineral content, and cation exchange capacity of samples collected at 12 random locations in the field. Soil bulk density, pH, clay fraction, soil organic C (SOC) content, and field slope data were Journal of Environmental Quality

then estimated corresponding to each soil property prediction for each 10-m pixel across the field (Ashtekar et al., 2014). To capture the spatial variability of soil properties and variable-rate N fertilization, DNDC was used to simulate 60 points in the field (n = 30 for each fertilization rate method). For sampling purposes, points on the 10-m gridded digital elevation model of the experimental field were divided into 10 groups, and for each N fertilization method three points from each group were randomly selected for simulation (Fig. 1). At each simulation location, site-specific inputs were drawn from the digital soil map and the measured yield and N rate data. Other model inputs, such as field management events and weather, were the same at all simulation locations. In each simulation, the model simulated a 30-yr spin-up period from 1979 to 2008 to stabilize soil C. Outputs from the last year of the simulations (i.e., 2008) are reported here, coinciding with the period of the experiment. The DNDC model was calibrated at each simulated field location using measured yield data and predicted SOC content from the digital soil map. The crop growth model was first calibrated to match measured yield by adjusting the optimum temperature for crop growth and maximum grain biomass production. Optimum temperature for crop growth was adjusted from its default value of 30 to 29°C. Each location simulated may have different maximum grain biomass production, and calibrated values ranged from 2820 to 5050 kg C ha-1 (the default value is 4123 kg C ha-1). At all locations the calibrated model grain yield prediction was within 1% of the observed yield. Soil organic C was calibrated by modifying initial organic C content at 0 to 10 cm to match simulated 2008 SOC to digital soil map values. Values of initial SOC used in calibration varied from 0.0172 to 0.02245 g C g-1 soil.

Statistical Analyses Yield and N loss data were tested for heteroscedasticity via the Fligner-Killeen tests (Crawley, 2007), and, if variance

assumptions were met, parametric tests were used. Producerdefined N application was compared with variable-rate N application with one-way ANOVA. When variance assumptions were violated, a nonparametric equivalent (Mann–Whitney– Wilcoxon U test) was used to test for differences between the populations of yield, gas emission, and N leaching data. Relationships between N application rate and yield, gas emission, and N leaching were tested with correlation analysis. To evaluate the differences in performance of variable-rate N fertilization relative to those of uniform-rate N fertilization, percent relative effect size was calculated as:

definition and scoping, inventory analysis, impact assessment, and interpretation. The goal of the LCA was to compare the life cycle environmental impacts of sensor-based, variable-rate N fertilization with those of fixed-rate N application. This study involved a comparative cradle-to-gate inventory, which is an accounting of resources used and emissions to the environment, from the extraction of natural resources for the production of inputs, to harvested corn grain at the edge of the field. The functional unit was taken to be 1 metric tonne (1 Mg) of harvested corn grain at a standard moisture content of 15.5% (wb). The Life Cycle Inventory (LCI) is an accounting of all inputs and outputs, as elementary flows, to and from the environment related to the full process of producing 1 Mg of corn grain. Primary LCI data were collected for corn production as practiced on the cooperator’s farm. Inputs included corn seed, fertilizer, herbicide, and fuel and energy consumed. Fuel and energy refers to the diesel fuel, gasoline, natural gas, electricity, or other on-farm energy consumption associated with farm operations within the system boundary. Background data for fuels used, emissions from the production of energy, and transportation were obtained from the Ecoinvent v3.1 database (Weidema et al., 2013). Primary foreground processes were all field operations: planting, herbicide application, two fertilizer applications (i.e., preplanting application and side-dress application), and harvesting. Diesel and lubricant consumption were calculated by assuming the use of typical equipment; determining power requirements; and estimating field capacity, efficiency, probable work days (Srivastava et al., 2006; Iowa State University Extension, 2002), and the associated fuel and oil use (ASABE, 2006a; ASABE, 2006b). Data sources for other inputs to the primary processes, including gasoline, liquefied petroleum gas, electricity, and natural gas, were taken from literature sources as detailed in Table 1. Farm chemical inputs use was as measured in the field experiment.

% Effect of Variable N Rate = (Variable N Rate – Producer N Rate/Producer N Rate) × 100 Data for each parameter of interest (i.e., yield, N application rate, ratio of applied N to yield, N2O emissions, NO3- leaching, and NH3 loss) were processed with 5000 iterations of a resampling with replacement simulation to calculate the mean percent effect size and 95% confidence interval. Effect size was considered significant if the confidence interval did not overlap 0. All statistical tests and the resample routine for effect size were performed in R and JMP v.11 (R Core Team, 2014; SAS Institute, 2013).

Life Cycle Assessment Life Cycle Assessment (LCA) has been recognized as an important framework for analyzing the environmental impacts of agricultural systems (Brentrup and Lammel, 2011; Cowell and Clift, 2000; Brentrup et al., 2004). The essence of LCA is the identification and evaluation of relevant environmental implications of a product, process, or system across its entire life span. By considering the entire lifecycle, LCA can avoid “problem shifting” between life cycle stages and receptors. Standard procedures for conducting LCAs have been developed by the USEPA, The Society of Environmental Toxicology and Chemistry, and the International Standards Organization (Fava et al., 1992; Vignon et al., 1993; ISO 2006). Life Cycle Assessment is commonly defined as a process comprising four stages: goal Table 1. Aggregated life cycle inputs to corn production. Inputs†

Variable rate 18.49 71.49 26.34 0.56 23.48 27.02

Producer rate 18.49 86.31 26.34 0.56 23.48 27.02

Lubricant, L ha-1

0.14

0.14

Gasoline, L ha-1 LPG, L ha-1 Electricity, MJ ha-1 Natural gas, m3 ha-1 Herbicide-embodied energy, MJ kg-1 Input transport energy, MJ kg-1 Input packaging–embodied energy, MJ ha-1 Custom work, MJ ha-1 Farm machinery–embodied energy, MJ ha-1

17.95 29.92 181.6 14.62 355.6 0.64 73.7 270 320

17.95 29.92 181.6 14.62 355.6 0.64 73.7 270 320

Urea, kg N ha-1 UAN 32%, kg N ha-1 DAP, kg N ha-1 Herbicide (glyphosate), kg active ingredient ha-1 Corn seed, kg N ha-1 Diesel, L ha-1

Source measured measured measured measured measured ASABE (2006a, 2006b), Srivastava et al. (2006), Iowa State University Extension (2002) ASABE (2006a, 2006b), Srivastava et al. (2006), Iowa State University Extension (2002) Shapouri et al. (2002) Shapouri et al. (2002) Shapouri et al. (2002) Shapouri et al. (2002) Farrell et al. (2006) Farrell et al. (2006) Graboski (2002) Shapouri et al. (2002) Graboski (2002)

† DAP, diammonium phosphate; LPG, liquefied petroleum gas; UAN, urea–ammonium–nitrate.

Journal of Environmental Quality

Many analyses of GHG emissions from agricultural systems report the flux of C to or from the soil. Whether a particular agricultural management will have a positive or negative impact on SOC, however, depends on local conditions that influence plant and soil processes driving SOC dynamics (Necpálová et al., 2014; Ogle et al., 2005) as well as the baseline level of SOC (Olson, 2010; Sanderman and Baldock, 2010). Because the yield differences between the two fertilization schemes examined here were extremely small, we anticipated no difference in change in soil organic matter or net soil CO2 flux between the two treatments. This hypothesis was verified by the DNDC model predictions showing no difference in soil C or soil CH4 fluxes between N treatments (results not shown) (one-way ANOVA; P > 0.90 in both cases). Therefore, because all management other than N application was identical in the two treatments and there was no predicted difference in C emissions, soil CO2 flux data are not reported, and N2O emissions are the only soil contributions to the GHG portion of the LCA. In most freshwater systems, enrichment of phosphorous (P) in the aquatic environment is the main driver of eutrophication (Khan and Mohammad, 2014). Phosphorous-driven eutrophication is commonly taken into account in LCA, but the best available methods are acknowledged to need improvement (Hauschild et al., 2013). Eutrophication from P is the result of P emissions reaching water bodies, which depend on a wide range of site-specific parameters, such as soil type, slope, slope length, soil cover, and proximity to water bodies. The most widely used LCA databases, such as the Ecoinvent database (Weidema et al., 2013), include estimates of average P emissions at the wholecountry or global scale and build on simplifying assumptions, such as constant P concentration in soil and fixed values for soil erodibility. These methods are not appropriate for this comparative LCA of N fertilization effects, and P-driven eutrophication is not accounted for in this analysis. The LCI data were translated into potential environmental impacts using Life Cycle Impact Assessment (LCIA) methods (ISO, 2006). The LCIA methods were selected following the recommended best practice of the International Reference Life Cycle Data System for midpoint methods (Hauschild et al., 2013). Selection of relevant impact categories was driven by the study goal of comparing the impacts of sensor-based, variable-rate N fertilization with those of fixed-rate N application. Because the only difference in management between the two treatments was N application, the impact categories selected are those related to the production and use of N fertilizer and the loss of N to air as N2O and to water as NO3-: cumulative energy consumption, GWP, and N-related eutrophication potential. Eutrophication potential from production of chemicals and fuels is expressed as units of NO3- equivalent. Eutrophication potential data were calculated using the Tool for the Reduction and Assessment of Chemical and Other Environmental Impacts using DNDC-predicted NO3- leaching loss (Bare, 2002).

Uncertainties and Limitations Although this analysis was performed with the best available information, uncertainties and limitations exist: background data taken from LCI databases and the literature may have different due to temporal and geographic scope; model-based predictions of soil biogeochemistry are limited by the accuracy of the Journal of Environmental Quality

model and are conditional on local conditions such as weather, soils, and management practices; and soil properties were based on extrapolations of in-field testing data and geomorphological inference that, although shown to be effective, are imperfect. On the other hand, structuring the analysis as a comparison reduces the importance of many of these uncertainties. Comparing the differences between the impacts of the two management practices effectively eliminates uncertainty and inaccuracy associated with inputs and unit processes that are common to both of the management scenarios. The response of the biogeochemical cycles to differences in N fertilization is representative of the local soils and climatic conditions observed during the field experiment, so caution is recommended when reinterpreting LCA results in a different context or for a different geographic location.

Results and Discussion The entire field received a preplant application of 45 kg N ha-1 of diammonium phosphate, so subsequent mention of fertilization rates refers only to post-plant side-dress amounts unless otherwise noted. Fertilization specified by the sensor-based algorithm resulted in lower N application than the producer-specified rate (Mann–Whitney–Wilcoxon test; U = 330, df = 30, P < 0.10), with mean N application rates of 71.5 and 86.3 kg N ha-1 for the variable- and fixed-rate applications, respectively. The majority of the variable-rate sample points were fertilized at the lower N rate limit requested by the cooperating farmer (56 kg N ha-1), and 11 points among the 30 sampled were at the highest rate (95 kg N ha-1). Another four points received N applications between the lowest and highest allowed by the variable-rate algorithm but were still lower than the fixed producer rate of 86 kg N ha-1 after the preplant application (Fig. 2A). The producer reported choosing the N rate limits imposed on the variable-rate system based on growing conditions at the time, the cost of fertilizer, and the historical yield data of the field. Although less N was applied using the sensor-based algorithm, the resulting yield was not different from that at the producer-defined N rate (one-way ANOVA; F1,58 = 0.01, P = 0.91) (Fig. 3; Table 2). Within the variable-rate sample points, however, yield was negatively correlated with N application rate due to eight low yield points that received the highest N rate (Fig. 2A). It is likely that the high N rate and low yield at these points are the result of less favorable soil properties, including lower organic matter and soil N contribution. Historically, this field was managed as several separate fields. The producer-defined and variable-rate N points with the lowest yield are located in the northeastern part of the field, which has been managed in row crops for an extended period, whereas the other parts of the field were in pasture or row crops alternating with orchard grass or alfalfa until the 1970s. Several of the points in the field that received the highest N rate under the sensor-based algorithm were near sites of visible erosion and sandy soil. At field locations where yield was lowest, the sensor-based N rate and resulting yield were higher than the producer-defined rate. At points where yield was more than 1 SD below the mean, the sensor-based N rate was on average 10% higher than the producer rate, and yield was on average 2.4% higher. Thus, there was a positive yield response to the increased N prescribed by the sensor.

Not surprisingly, all forms of N loss were positively correlated with total N application rate, irrespective of how the rate was specified. Within the variable-rate sample points, the highest losses occurred at locations with the highest N application rate and the lowest grain yield. Locations where the sensor-prescribed N applications exceeded the producer-defined N rate are also the locations of the highest N losses (Fig. 2B, C). Although reactive N losses were lower with sensor-based fertilization, there was also significantly higher variability in the N losses (Table 2; Fig. 2B, C). As seen in Fig. 2, at all but a few points, the sensor-based algorithm specified N at either the highest or lowest allowed rate, resulting in a bimodal distribution of N application rate. The bimodal distribution of N application rates specified by the sensor-based algorithm results in bimodal distributions of N2O flux and NO3- loss, as shown in Fig. 2B and 2C. The higher standard deviations of N losses under variablerate N relative to fixed-rate N reported in Table 2 are the result of these two modes being relatively far apart rather than the existence of large variation around each mode. For example, taken separately, the SD of N2O emissions at points that received 56 kg N ha-1 (SD, 0.20 kg N ha-1) is smaller than that of the points that received the fixed N rate (SD, 0.32 kg N ha-1) and is the same as those that received the highest N rate.

Life Cycle Impact Assessment

Fig. 2. Comparison of N fertilization rates and corresponding corn yield (A), N2O emissions (B), and NO3− leaching (C).

At field locations where N was applied at the lower limit requested by the farmer, it is possible that even lower application rates would have maintained yield, with even greater overall return to N. Even with the somewhat narrow limits imposed by the cooperating farmer, as shown in Fig. 3, there was a significantly higher ratio of yield per unit N in the variable-rate strips (35.1) relative to the producer-defined N rate (29.8) strips (U = 275, df = 30, P < 0.02). Estimation of losses of N to the environment using DNDC suggest that variable-rate fertilization significantly decreased losses of N to both the atmosphere and to water (Fig. 3). Mean annual N2O emissions were lower under variable-rate N application (U = 283.5, df = 30, P < 0.02), as were NH3 losses (U = 277.5, df = 30, P < 0.05). Leaching of NO3- at the sampled points was lower with variable-rate N application, but the effect was more variable (Wilcoxon rank sum test; W = 577, P = 0.06).

Total life cycle nonrenewable energy use was lower under sensor-based variable N application than the fixed producer rate (Table 3). The variable N rate required 1273 MJ of energy per megagram of corn produced, compared with 1371 MJ per megagram corn with the producer-defined N rate. The largest component of energy use for both fertilizer application methods, and the most significant difference between them, is the manufacture of synthetic N fertilizer, which accounts for 58 and 61% of total energy for the variable N and producer-defined application rates, respectively. Other types of energy use included transportation of chemical inputs, commodity packaging, transportation, and machinery operation. Because fertilizer rates differed between the treatments, energy consumption for transportation of chemical inputs was 9 MJ Mg-1 with variable-rate application, compared with 10 MJ Mg-1 required for the producer rate. Diesel used in corn production was approximately 118 MJ Mg-1 in both treatments because machinery use was the same regardless of what amount of fertilizer was applied, and corn yield was nearly the same. Variable-rate N fertilization emitted 10% fewer greenhouse gases than the producer-defined N rate application (405 vs. 450 kg CO2 eq Mg-1 corn) (Table 3). Soil N2O emission was the largest component of GWP in both treatments, contributing 76% of the total in both cases. Upstream GHG emissions resulting from fertilizer production was 14% lower in the variable-rate case and contributed 15% of total GWP in both rate treatments. The LCIA indicated that N-related eutrophication potential was approximately 16% lower with variable-rate N application relative to the fixed rate (6.4 vs. 7.6 kg NO3 eq) (Table 3). The vast majority of eutrophication potential related to N use was the result of NO3- losses from the field (97% in both treatments). Life cycle acidification potential was 22% lower in the variable-rate N system than in the producer-defined N rate system Journal of Environmental Quality

Fig. 3. Percent effect size comparing variable-rate N fertilizer application and producer-defined application rate for yield and N loss parameters for a corn system in Lincoln County, MO. Black points represent mean percent effect of variable rate compared with producer-defined N application rate ± 95% confidence interval calculated after resampling with replacement (1000 iterations) of the 60 application points (n = 30 for both variable and producer rate).

(1.6 vs. 2.0 kg SO2 eq) (Table 3). Soil NH3 emission was the largest source of acidification potential in both systems, contributing 85 and 87% in the variable-rate and fixed-rate systems, respectively. Fertilizer production was the next largest component of acidification potential, contributing approximately 7% in both systems.

Future Work and Conclusions Our analysis, incorporating direct measurements, model simulation, and a LCA, suggest that, relative to a uniform rate of fertilizer application, corn production using a sensor-based, variable-rate N application system can significantly decrease both gaseous and aqueous N losses. The model can be more fully validated with continued field data collection. However, the strong correlation between DNDC yield projections and measured harvest suggests that use of a crop canopy sensor and variablerate N application reduced total N fertilizer use by 11% with no significant reduction in grain yield compared with uniform fertilization and that variable-rate fertilization had lower life cycle impacts because sensor-based N application required 7% less total energy per ton of corn. Furthermore, the DNDC simulation results suggest that variable-rate fertilization mitigated soil N2O emissions and volatilized NH3 loss and NO3- leaching by 10, 23, and 16%, respectively. Including emissions associated with producing farm inputs, variable-rate N management also resulted in 10% less GWP, 22% less acidification potential, and 16% less eutrophication potential than the producer-specified N rate. The current analysis compared one particular sensor-based fertilization algorithm with a single fixed-rate of fertilization chosen by the cooperating farmer. The weather and soil Table 2. Comparison of measured yield, measured N application rate, model-predicted N2O emission, and model-predicted NO3- leaching between sensor-based variable N application and producer-defined (uniform) N application, Lincoln County, MO. Measurement Yield, kg C ha-1 Application rate, kg N ha-1 N2O emissions, kg N ha-1 NO3- leaching, kg N ha-1 NH3 emissions, kg N ha-1

Producer N application

Variable N application

3916 (614.8)† 86.3 (0.0) 7.04 (0.32) 72.1 (20.4) 9.33 (1.06)

3932 (527.3) 71.5 (18.7) 6.34 (0.88) 60.8 (31.2) 7.14 (2.82)

† Values are means with SD in parentheses. Journal of Environmental Quality

conditions simulated are representative of the local climate and soil conditions during the field experiment. It is likely that the trends in relative performance observed will translate to other years and locations because of the commonality of the governing Table 3. Life cycle impact indicators of producer-defined and sensorbased variable N fertilizer application with significant contributions by life cycle stage and total. Life cycle impact

Life cycle stage

Nonrenewable energy use fertilizer production diesel fuel production corn seed production gasoline production liquefied petroleum gas production natural gas production electricity production herbicide production other energy use total Global warming potential soil N2O emission fertilizer production diesel fuel production corn seed production gasoline production liquefied petroleum gas production natural gas production electricity production herbicide production other energy use total Eutrophication potential soil NO3 leaching corn seed production fertilizer production total Acidification potential soil NH3 emissions fertilizer production total

Producer N rate

Variable N rate

——— MJ Mg-1 corn ——— 815 720 118 117 83 83 74 73 73

73

58 51 20 78 1371

58 51 20 77 1273

— kg CO2 eq Mg-1 corn — 342 307 69 60 10 10 5 5 6 6 7 7 4 1 1 6 450

4 1 1 6 405

—— kg NO3 eq Mg-1 corn — 7.4 6.2 0.1 0.1 0.1 0.1 7.6 6.4 — kg SO2 eq Mg-1 corn — 1.8 1.4 0.2 0.2 2.0 1.6

biogeochemical processes and the large performance advantages we estimate for the sensor-based, variable-rate system. In fact, the relative performance of the sensor-based system could have been better in this analysis if the lower fertilization limit imposed by the cooperating farmer had been relaxed. It is clear that N use efficiency and environmental performance can be improved if better sensor-based algorithms can be developed that can more accurately provide N at efficient rates. By maintaining yields and mitigating environmentally costly N losses, sensor-based variable N application technology has the potential to be a powerful tool in sustainable precision agriculture.

Acknowledgments This research is part of a regional collaborative project supported by the USDA–NIFA, Award No. 2011-68002-30190, “Cropping Systems Coordinated Agricultural Project: Climate Change, Mitigation, and Adaptation in Corn-based Cropping Systems.” Project Web site: sustainablecorn.org.

References American Society of Agricultural and Biological Engineers (ASABE). 2006a. ASABE Standards EP. 496.3: Agricultural machinery management. American Society of Agricultural and Biological Engineers, St. Joseph, MI. American Society of Agricultural and Biological Engineers (ASABE). 2006b. ASABE Standards EP. 497.7: Agricultural machinery management data. American Society of Agricultural and Biological Engineers, St. Joseph, MI. Aneja, V.P., W.H. Schlesinger, and J.W. Erisman. 2009. Effects of agriculture upon the air quality and climate: Research, policy, and regulations. Environ. Sci. Technol. 43:4234–4240. doi:10.1021/es8024403 Ashtekar, J.M., P.R. Owens, R.A. Brown, H.E. Winzeler, M. Dorantes, Z. Libohova, M. Dasilva, and A. Castro. 2014. Digital mapping of soil properties and associated uncertainties in the Llanos Orientales, South America. In: D. Arrouays, N. McKenzie, J. Hempel, A.R. Forges, and A. McBratney, editors, GlobalSoilMap: Basis of the global spatial soil information system. Taylor and Francis Group, London. p. 367. Barker, D.W., and J.E. Sawyer. 2010. Using active canopy sensors to quantify corn nitrogen stress and nitrogen application rate. Agron. J. 102:964–971. doi:10.2134/agronj2010.0004 Bare, J.C. 2002. Traci. J. Ind. Ecol. 6:49–78. doi:10.1162/108819802766269539 Brentrup, F., and J. Lammel. 2011. LCA to assess the environmental impact of different fertilisers and agricultural systems. International Fertiliser Society, York, UK. Brentrup, F., J. Küsters, J. Lammel, P. Barraclough, and H. Kuhlmann. 2004. Environmental impact assessment of agricultural production systems using the life cycle assessment (LCA) methodology II: The application to N fertilizer use in winter wheat production systems. Eur. J. Agron. 20:265–279. doi:10.1016/S1161-0301(03)00039-X Cassman, K.G., A. Dobermann, and D.T. Walters. 2002. Agroecosystems, nitrogen-use efficiency, and nitrogen management. Ambio 31:132–140. doi:10.1579/0044-7447-31.2.132 Climate Action Reserve. 2013. Nitrogen management project protocol. Version 1.1. Climate Action Reserve, Los Angeles, CA. Cowell, S.J., and R. Clift. 2000. A methodology for assessing soil quantity and quality in life cycle assessment. J. Clean. Prod. 8:321–331. doi:10.1016/ S0959-6526(00)00023-8 Crawley, M.J. 2007. The R book. John Wiley & Sons, West Sussex, UK. Dellinger, A.E., J.P. Schmidt, and D.B. Beegle. 2008. Developing nitrogen fertilizer recommendations for corn using an active sensor. Agron. J. 100:1546– 1552. doi:10.2134/agronj2007.0386 FAOSTAT (Food and Agriculture Organization of the United Nations). 2014. Crop production data. Food and Agriculture Organization of the United Nations. http://faostat.fao.org/site/339/default.aspx (accessed 22 May 2014). Farrell, A.E., R.J. Plevin, B.T. Turner, A.D. Jones, M. O’Hare, and D.M. Kammen. 2006. Ethanol can contribute to energy and environmental goals. Science 311:506–508. Fava, J.A., R.F. Weston, F. Consoli, R. Denison, K. Dickson, T. Mohin, and B. Vignon. 1992. A conceptual framework for life-cycle impact assessment. Society of Environmental Toxicology and Chemistry, Pensacola, FL.

Galloway, J.N., J.D. Aber, J.W. Erisman, S.P. Seitzinger, R.W. Howarth, E.B. Cowling, and B.J. Cosby. 2003. The nitrogen cascade. Bioscience 53:341– 356. doi:10.1641/0006-3568(2003)053[0341:TNC]2.0.CO;2 Graboski, M.S. 2002. Fossil energy use in the manufacture of corn ethanol. National Corn Growers Association, Washington, DC. Gruber, N., and J.N. Galloway. 2008. An earth-system perspective of the global nitrogen cycle. Nature 451:293–296. doi:10.1038/nature06592 Hauschild, M.Z., M. Goedkoop, J. Guinée, R. Heijungs, M. Huijbregts, O. Jolliet, M. Margini, A. De Schryver, S. Humbert, A. Laurent, S. Sala, and R. Pant. 2013. Identifying best existing practice for characterization modeling in life cycle impact assessment. Int. J. Life Cycle Assess. 18:683–697. doi:10.1007/s11367-012-0489-5 Iowa State University Extension. 2002. Estimating the field capacity of farm machines. Extension Publication PM 696; Ag Decision Maker, File A324. www.extension.iastate.edu/agdm/crops/html/a3-24.html (accessed 5 Nov. 2014). ISO. 2006. ISO 14040: Environmental management-life cycle assessment-principles and framework. International Organization for Standardization, Geneva. Johnston, A.M., and T.W. Bruulsema. 2014. 4R nutrient stewardship for improved nutrient use efficiency. Procedia Eng. 83:365–370. doi:10.1016/j. proeng.2014.09.029 Khan, M.N., and F. Mohammad. 2014. Eutrophication: Challenges and solutions. In: A.A. Ansari and S.S. Gill, editors, Eutrophication: Causes, consequences and control. Springer, The Netherlands. p. 1–15. Kitchen, N.R., K.A. Sudduth, S.T. Drummond, P.C. Scharf, H.L. Palm, D.F. Roberts, and E.D. Vories. 2010. Ground-based canopy reflectance sensing for variable-rate nitrogen corn fertilization. Agron. J. 102:71–84. doi:10.2134/agronj2009.0114 Li, C., N. Farahbakhshazad, D.B. Jaynes, D.L. Dinnes, W. Salas, and D. McLaughlin. 2006. Modeling nitrate leaching with a biogeochemical model modified based on observations in a row-crop field in Iowa. Ecol. Modell. 196:116–130. doi:10.1016/j.ecolmodel.2006.02.007 Li, C., S. Frolking, G.J. Crocker, P.R. Grace, J. Klír, M. Körchens, and P.R. Poulton. 1997. Simulating trends in soil organic carbon in long-term experiments using the DNDC model. Geoderma 81:45–60. doi:10.1016/ S0016-7061(97)00080-3 Li, C., S. Frolking, and T.A. Frolking. 1992. A model of nitrous oxide evolution from soil driven by rainfall events: 2. Model applications. J. Geophys. Res. 97:9777–9783. doi:10.1029/92JD00510 Mamo, M., G.L. Malzer, D.J. Mulla, D.R. Huggins, and J. Strock. 2003. Spatial and temporal variation in economically optimum nitrogen rate for corn. Agron. J. 95:958–964. doi:10.2134/agronj2003.0958 McLoud, P.R., R. Gronwald, and H. Kuykendall. 2007. Precision agriculture: NRCS Support for Emerging Technologies. Agronomy Technical Note No. 1. USDA–NRCS, Washington, DC. Millar, N., G.P. Robertson, A. Diamant, R.J. Gehl, P.R. Grace, and J.P. Hoben. 2013. Quantifying N2O emissions reductions in agricultural crops trough nitrogen fertilizer rate reduction. Verified Cabron Standard, Washington. www.v-c-s.org/methodologies/quantifying-n2o-emissions-reductionsagricultural-crops-through-nitrogen-fertilizer (accessed 17 June 2015). Millar, N., G.P. Robertson, A. Diamant, R.J. Gehl, P.R. Grace, and J.P. Hoben. 2012. Methodology for quantifying nitrous oxide (N2O) emissions reductions by reducing nitrogen fertilizer use on agricultural crops. American Carbon Registry, Winrock Int., Little Rock, AR. Nash, E., P. Korduan, and R. Bill. 2009. Applications of open geospatial web services in precision agriculture: A review. Precis. Agric. 10:546–560. doi:10.1007/s11119-009-9134-0 NOAA. 2014. National Weather Service Cooperative observer program. NOAA Cooperative Network data obtained from the Iowa Environmental Mesonet. http://mesonet.agron.iastate.edu/request/coop/ fe.phtml?network=MOCLIMATE (accessed 1 June 2014). Necpálová, M., R.P. Anex, A.N. Kravchenko, L.J. Abendroth, S.J. Del Grosso, W.A. Dick, M.J. Helmers, D. Herzmann, J.G. Lauer, E.D. Nafziger, J.E. Sawyer, P.C. Scharf, J.S. Strock, and M.B. Villamil. 2014. What does it take to detect a change in soil carbon stock? A regional comparison of minimum detectable difference and experiment duration in the north central United States. J. Soil Water Conserv. 69:517–531. doi:10.2489/jswc.69.6.517 Ogle, S.M., F.J. Breidt, and K. Paustian. 2005. Agricultural management impacts on soil organic carbon storage under moist and dry climatic conditions of temperate and tropical regions. Biogeochemistry 72:87–121. doi:10.1007/ s10533-004-0360-2 Olson, K.R. 2010. Impacts of tillage, slope, and erosion on soil organic carbon retention. Soil Sci. 175:562–567. doi:10.1097/SS.0b013e3181fa2837 R Core Team. 2014. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Journal of Environmental Quality

Ravishankara, A.R., J.S. Daniel, and R.W. Portmann. 2009. Nitrous oxide (N2O): The dominant ozone-depleting substance emitted in the 21st Century. Science 326:123–125. doi:10.1126/science.1176985 Reay, D.S., E.A. Davidson, K.A. Smith, P. Smith, J.M. Melillo, F. Dentener, and P.J. Crutzen. 2012. Global agriculture and nitrous oxide emissions Nature Clim. Change 2:410–416. Roberts, D.F., N.R. Kitchen, P.C. Scharf, and K.A. Sudduth. 2010. Will variablerate nitrogen fertilization using corn canopy reflectance sensing deliver environmental benefits? Agron. J. 102:85–95. doi:10.2134/agronj2009.0115 Sanderman, J., and J.A. Baldock. 2010. Accounting for soil carbon sequestration in national inventories: A soil scientist’s perspective. Environ. Res. Lett. 5:034003. doi:10.1088/1748-9326/5/3/034003 SAS Institute. 2013. Using JMP 11. SAS Institute Inc., Cary, NC. Scharf, P.C., N.R. Kitchen, K.A. Sudduth, J.G. Davis, V.C. Hubbard, and J.A. Lory. 2005. Field-scale variability in optimal nitrogen fertilizer rate for corn. Agron. J. 97:452–461. doi:10.2134/agronj2005.0452 Scharf, P.C., and J.A. Lory. 2009. Calibrating reflectance measurements to predict optimal sidedress nitrogen rate for corn. Agron. J. 101:615–625. doi:10.2134/agronj2008.0111 Scharf, P.C., D.K. Shannon, H.L. Palm, K.A. Sudduth, S.T. Drummond, N.R. Kitchen, L.J. Mueller, V.C. Hubbard, and L.F. Oliveira. 2011. Sensorbased nitrogen applications out-performed producer-chosen rates for corn in on-farm demonstrations. Agron. J. 103:1683–1691. doi:10.2134/ agronj2011.0164 Schlesinger, W.H. 1997. Biogeochemsitry. Academic Press, San Diego, CA. Schmidt, J.P., A.E. Dellinger, and D.B. Beegle. 2009. Nitrogen recommendations for corn: An on-the-go sensor compared with current recommendation methods. Agron. J. 101:916–924. doi:10.2134/agronj2008.0231x Shanahan, J.F., N.R. Kitchen, W.R. Raun, and J.S. Schepers. 2008. Responsive in-season nitrogen management for cereals. Comput. Electron. Agric. 61:51–62. doi:10.1016/j.compag.2007.06.006 Shapouri, H., J. Duffield, and M. Wang. 2002. The energy balance of corn ethanol: An update. AER813. Office of the Chief Economist, Office of Energy Policy and New Uses, USDA, Washington, DC. https://anl.app.box. com/s/1g9qrss23gixy5y8ohj8oebxjftsyjl8 (accessed 7 Jan. 2016). Shaver, T.M., R. Khosla, and D.G. Westfall. 2011. Evaluation of two crop canopy sensors for nitrogen variability determination in irrigated maize. Precis. Agric. 12:892–904. doi:10.1007/s11119-011-9229-2

Journal of Environmental Quality

Smith, W.N., R.L. Desjardins, B. Grant, C. Li, R. Lemke, P. Rochette, M.D. Corre, and D. Pennock. 2002. Testing the DNDC model using N2O emissions at two experimental sites in Canada. Can. J. Soil Sci. 82:365–374. doi:10.4141/S01-048 Smith, W.N., B.B. Grant, R.L. Desjardins, P. Rochette, C.F. Drury, and C. Li. 2008. Evaluation of two process-based models to estimate soil N2O emissions in Eastern Canada. Can. J. Soil Sci. 88:251–260. doi:10.4141/ CJSS06030 Snyder, C.S. 2012. Are Midwest corn farmers over-applying fertilizer N? Better Crops Plant Food 96:3–4. Srivastava, A.K., C.E. Goering, R.P. Rohrbach, and D.R. Buckmaster. 2006. Machinery selection and management. In: Engineering principles of agricultural machines, 2nd ed. ASABE, St. Joseph, MI. p. 525–552. Tonitto, C., M.B. David, L.E. Drinkwater, and C. Li. 2007. Application of the DNDC model to tile-drained Illinois agroecosystems: Model calibration, validation, and uncertainty analysis. Nutr. Cycling Agroecosyst. 78:51–63. doi:10.1007/s10705-006-9076-0 USDA–NASS. 2014. Data and statistics: Quick Stats 2.0. USDA National Agricultural Statistics Service. Washington, DC. http://quickstats.nass.usda. gov (accessed 2 June 2014). Vanotti, M.B., and L.G. Bundy. 1994. Corn nitrogen recommendations based on yield response data. J. Prod. Agric. 7:249–256. doi:10.2134/jpa1994.0249 Vignon, B.W., D.A. Tolle, B.W. Cornaby, H.C. Latham, C.L. Harrison, T.L. Boguski, R.G. Hunt, and J.D. Sellers. 1993. Life-cycle assessment: Inventory guidelines and principles (EPA/600/R-92/245). USEPA, Cincinnati, OH. Walburg, G., M.E. Bauer, and C.S.T. Daughtry. 1982. Effects of nitrogen nutrition on the growth, yield and reflectance characteristics of corn canopies. Agron. J. 74:677–683. doi:10.2134/agronj1982.0002196200740004002 0x Weidema, B.P., C. Bauer, R. Hischier, C. Mutel, T. Nemecek, J. Reinhard, C.O. Vadenbo, and G. Wernet. 2013. Overview and methodology: Data quality guideline for the Ecoinvent database version 3. Swiss Centre for Life Cycle Inventories, Zurich, Switzerland.