Hot spots of opportunity for improved cropland nitrogen management across the United States

We obtained county-level N flows for the contiguous US in 2011, 2012, and 2013, including farm fertilizer N, recoverable manure N, N fixation by legumes, and N in crop harvests, from the NuGIS database [20]. Cropland N balance is calculated in NuGIS as:

$$\begin{align*} \text{N}_{\text{balance}} & \,{=}\,\left( {{\text{N}}_{{\text{farm}{\_}\text{fertilizer}}}}{\text{ + }}{{\text{N}}_{{\text{recoverable}{\_}\text{manure}}}}{\text{ + }}{{\text{N}}_{{\text{fixation}}}} \!\right)\\ & \quad - {{\text{N}}_{{\text{crop}{\_}\text{harvest}}}}\end{align*}$$

This is a partial N mass balance because it does not account for a number of additional N flows including atmospheric deposition, nutrients in irrigation water, land application of biosolids, or several pathways of N loss to the environment, such as eroded soil, gaseous N emissions, and leaching [20]. Previous studies have used this calculation for partial cropland N balance as well [21, 30]. 'Recoverable' manure N represents the amount of N in excreted manure that would be available to apply to cropland and does not include N lost during manure storage and handling [20]. The areal version of the mass balance is calculated by dividing N_balance by the total cropland area for each county. County-level N use efficiency (NUE, %) is calculated as:

$$\begin{align*}{\text{NUE}} & = \left( {{\text{N}}_{{\text{crop}{\_}\text{harvest}}}}{{/}}\left( {{\text{N}}_{{\text{farm}{\_}\text{fertilizer}}}} \right.\right.\\ &\quad\left.\left.{\text{ + }}{{\text{N}}_{{\text{recoverable}{\_}\text{manure}}}}{\text{ + }}{{\text{N}}_{{\text{fixation}}}} \right) \right){ } \times {{ 100 }}\end{align*}$$

We provide more discussion on the N balance methodology, including its limitations, in the supplementary materials.

We collected county-level data spanning government program participation, farm economics, population demographics, climate change belief and policy support, biophysical characteristics, and crop type as potential predictors of areal cropland N balance (table S1 (available online at stacks.iop.org/ERL/16/035004/mmedia)). All data were for the year 2012 or the 2011–2013 annual average, with the exception of population density (2010) and climate change belief and policy support (2016). In some cases, data filling and manipulation were required as described in the supplementary materials. We also created a cropland typology for 2012 based on county-level crop land areas and consisting of ten categorical variables (i.e. typology groups), which we described in detail in a previously published data note [29] and summarize here in table 1 and figure 1.

Zoom In Zoom Out Reset image size

We used the 'segmented' package in R [31, 32] to determine breakpoints in segmented linear regression for county-level N harvested in crops versus county-level total N inputs, using 2011–2013 means [20]. We identified one breakpoint for all counties included in the study, as well as for individual cropland typology groups as a whole (i.e. not based on individual crops) and determined multiple related statistics, including the standard error and P-value for the breakpoint, as well as slopes, r², and P-values for the linear correlations below and above the breakpoints. 'Total N input' here includes N fixation by legumes (N_input = N_{farm_fertilizer} + N_{recoverable_manure} + N_fixation), which is important when considering N balances. Note that the results of this analysis do not represent any single crop type, but rather the groups of counties having similar crop mixes according to the typology ([29]; table 1).

To analyze factors correlated with mean N balance (kg N ha⁻¹ yr⁻¹) in 2011–2013, we ran a series of stepwise hierarchical random effects models. We used stepwise models because of the large number of potential independent variables. First, we ran five separate models to predict county-level areal N balance, including models using the following variable types (a) government program participation; (b) farm economics; (c) population demographics; (d) biophysical attributes; and (e) cropland typology groups. Stepwise models included a random effect at the state level, to account for factors that may influence N balance within a given state (e.g. state policies) that are not captured in the model. Statistically significant variables from the stepwise models (P < 0.05), as well as the aggregated climate scale variable, were included in a final hierarchical random effects model, also with a random effect at the state level.

Our spatial analysis included a combination of criteria scoring and hot spot analysis, as described in figure 2 and the supplementary materials. We qualified 20 hot spots of mostly contiguous counties based on the sum of three criteria scores (Total N Surplus, Excessive N Input, Potential for Improvement; figure 2), including the majority of high scoring counties. Finally, we aggregated data for these hot spots, including total cropland and different N flows in metric tons, and calculated the areal N balance and the ratio of N_{farm_fertilizer} to N_{recoverable_manure} for each hot spot.

Zoom In Zoom Out Reset image size

During 2011–2013, the 2887 US counties included in our study were characterized by a mean overall N balance of +4.0 million metric tons N per year (MMT N yr⁻¹) across all croplands, or +25 kg N ha⁻¹ yr⁻¹, and an overall N recovery in crops of 79%. Cumulative mean annual cropland N flows for that period included 11.6 MMT N yr⁻¹ in farm fertilizer, 1.05 MMT N yr⁻¹ in recoverable animal manure, 6.0 MMT N yr⁻¹ via N fixation by legumes, and 14.7 MMT N yr⁻¹ harvested in crops for 157.3 million ha of total cropland. County-level N balances per hectare of cropland (i.e. areal N balance) during 2011–2013 varied greatly across the US, as did the total N balance per county (figures 3(a) and (b)) and NUE (figure S1) [20].

Zoom In Zoom Out Reset image size

Considering all counties, the input and output components of the N balance were strongly correlated up to a breakpoint in N input (mean ± std. error = 175 ± 3 kg N ha⁻¹ yr⁻¹, P< 0.001) with peak crop N yield response, after which increased total N input was negatively correlated with N output in crop yields (figure 4(a), table S2). Therefore, we assume that additional N input beyond the breakpoint is more likely to be lost to the environment. We further identify breakpoints in N input for counties within different cropland typology groups (figure 1). We find significant breakpoints in total N input for eight out of the ten cropland typology groups ranging from 83 ± 11 to 230 ± 5 kg N ha⁻¹ yr⁻¹ (P < 0.02 in all cases; figure 4(b), table S2). For approximately 25% of counties in our analysis, total N inputs exceeded the most appropriate breakpoint value (see section 2), which indicates excessive N use beyond the levels associated with peak yield response (figure S2).

Zoom In Zoom Out Reset image size

We find multiple factors across a number of different categories predict N balance (table 2). Greater N balance was associated with counties having farms with greater total operating expenses (P < 0.001), greater population density (P < 0.001), lesser climate change belief and policy support ('climate-scale') (P = 0.022), and lesser soil productivity (P = 0.010), as well as, to a weaker extent, greater precipitation (P = 0.055) and lesser participation in other federal USDA programs (i.e. federal programs excluding the Conservation Reserve Program (CRP), Wetlands Reserve Program (WRP), Farmable Wetlands Program (FWP), and Conservation Reserve Enhancement Program (CREP)) (P = 0.075). We also find that a number of cropland typology groups (table 1) correlate with greater N balance, including groups 1 (P = 0.010), 3 (P = 0.033), 4 (P = 0.001), and 9 (P = 0.029).

Twenty hot spots emerged in our analysis, including 759 counties across the West, Midwest, and South (figure 5). These hot spots hosted ∼24% of the cropland area included in our study, but accounted for ∼63% of the total N surplus (table 3). The overall areal N mass balance rate for hot spot counties (+68 kg N ha⁻¹ yr⁻¹) was 2.7 times greater than the overall rate for all counties. Overall NUE (% of N inputs recovered in harvested crops) in hot spots ranged from 17% to 75% (table 3), indicating that while some hot spots are characterized by markedly inefficient use of N per hectare of cropland, others hosted substantial tonnage of surplus N use despite more efficient N use per hectare due to their large total cropland area. N input metrics for each hot spot, including fertilizer, recoverable manure, and N fixation by legumes, indicate that the importance of different inputs in N balances varies greatly across hot spots (table 3). For example, fertilizer to manure input ratios on an N basis ranged from 1 to >100. Additionally, N fixation by legumes was a sizeable reactive N input, accounting for 26% of N inputs for all hot spots combined, but was of differing importance in N balances across hot spots (table 3). Hot spots accounted for 38% and 49% of overall fertilizer and manure inputs to US croplands, respectively.

Zoom In Zoom Out Reset image size

Our findings contribute several important advancements in the context of current literature. First, we provide new insights into the relationship between cropland N inputs and outputs in the US using county-level data, including breakpoint values that can identify excessive N use (figure 4). Data aggregated by state in 2011–2013 (figure S3) or at the national level over time [7] do not to reveal breakpoints in the relationship between N inputs and outputs for US croplands. Furthermore, our use of a new cropland typology [29] (figure 1, table 1) allows us to better associate county-level N balances, as well as breakpoints in the relationship between cropland N input and output, with specific dominant crop mixes across the US for our study period (figure 4(b)). Previous investigators have been unable to assess N balance in this way, instead focusing on USDA ERS Farm Resource Regions based on farming characteristics in the 1990s [22]. Crop-specific analysis for the US has identified N fertilizer input above which corn yields plateau (150 kg N ha⁻¹ yr⁻¹) and at which peak yield occurs for winter wheat (50 kg N ha⁻¹ yr⁻¹), but spatial resolution is limited to the state scale [33]. Our analysis of county-level data in figure 4 is analogous to the economically optimal N rate (EONR) approach commonly used at the farm- or field-scale [34]. Like EONR analyses, we observe diminishing returns beyond the breakpoint associated with peak crop N yield response in the model including all counties, as well as in the model for cropland typology group 9 (dominated by corn grain and soybeans) (table S2).

Second, our results highlight diverse factors that are significantly correlated with areal N balance for US croplands, including total farm operating expenses, population density, climate change belief and policy support, inherent soil productivity, and crop typology groups (table 2). Our analysis is unique in its integration of disciplinary perspectives and can inform future N modeling and policy efforts, including guidance for considering specific populations. Additionally, we highlight counties with greater N balance than predicted based on the factors in our model, which could potentially be areas of focus more capable of shifting N practices (figure 5(c)). Previous efforts to model N flows, budgets, and mitigation in the US have generally not included social, demographic, and economic factors that can influence N management [8, 35, 36]. Simultaneously, analyses to understand the social and economic factors that predict N management generally have not been linked to biophysical data assessing whether such behaviors influence environmental outcomes [26].

Third, our use of spatial clustering methods to assess nutrient flows and balances across large spatial scales has few precedents in the literature [37, 38]. Spatial targeting is expected to improve the cost-effectiveness of agri-environmental policy because applying conservation measures in the most suitable regions can provide environmental benefits at lower costs [39, 40]. Additionally, spatial targeting of N policy and programs can likely help alleviate some of the major challenges faced by producers transitioning to new N management practices, if it involves the creation of new networks of farm advisors along with new infrastructure and technologies needed to overcome path dependency associated with habituated models of N management [41]. Finally, there is some evidence that spatial targeting of agri-environmental policy provides neighborhood effects such as higher levels of social acceptability [42].

Our results have several implications for future N management, policy, and programs. First, our analysis can inform efforts to increase uptake of existing, and currently underutilized, voluntary carbon offset programs for N management, specifically incentive programs including efforts to pay producers to reduce their N fertilizer use [43–45]. The adoption of these programs is very low, in part because of concerns over negative yield impacts from N input reductions. Evidence suggests that producers are much more likely to support adoption of N use efficiency practices, rather than N input reduction practices, likely because of the potential yield implications of N input reduction [46]. The discrepancy for the Midwest between figures 5(a) and (b) lends some support to this concern. While numerous Midwest counties score high for the Total Surplus N criterion (due to large cropping areas), most do not score high for the Excessive N Input criterion. This is because their county-level N inputs per hectare, while relatively high, are still below the N input breakpoint beyond which N in crop harvests plateaus or declines (figure S2). Our analysis also suggests that, for other counties and regions (largely in the South and West), N use efficiency and N input reduction strategies are likely one in the same, providing environmental gains without necessarily risking losses in terms of crop yield, farm profits, and food security (figures 5(b), S1 and S2). This finding may be critical for producer acceptance because it could suggest minimal economic losses or even increased profitability [34]. While continued focus on the Mississippi River watershed is necessary due to the scale of agriculture, associated N flows (figure 3(b)), and environmental impact [10, 47], improvements in cropland N cycling on a per hectare basis may be more easily achieved elsewhere in the US. However, uptake of protocols in these regions may require additional field data to parameterize models of given crop types by region for location-specific understanding of the impact of N input reductions on yields, profitability, and environmental outcomes [45].

Our analysis also highlights important crop types to consider in N policy efforts, and illuminates the relative importance of fertilizer N versus manure N by location. The three most common cropland typology groups for hot spot counties (in descending order) were groups 3, 9, and 1, which include substantial areas of hay, corn grain, soybeans, wheat, and other crops (tables 1 and 3). Thus, many of the hot spots in figure 5(e) are dominated by animal feed crops, highlighting the link between food system trends toward meat consumption and N dynamics [48]. In hot spots where the fertilizer:manure N input ratio is high (e.g. >10), many counties are producing animal feed destined for animals located outside the county or state, and local manure availability is limited [49]. Conversely, in regions where the fertilizer:manure N input ratio is relatively low (e.g. <10), manure management should be an especially critical focus, including ways to optimize its use as an N source to croplands on its own and in combination with fertilizer and/or biological N fixation [49, 50]. From a policy standpoint, this is a critical distinction, because relevant government programs to improve N fertilizer and manure management vary in their focus; for example, the Environmental Quality Incentives Program (EQIP) mandates 60% of funding be utilized for livestock projects, especially relevant where animal agriculture is prevalent and characterized by excess manure N. Conversely, other programs may be most relevant in regions where synthetic fertilizer N is more dominant. Appropriate scaling of co-located cropping and livestock systems, whether neighboring or fully integrated, is one approach that has potential to reduce the need for imported N-containing feed and fertilizer while facilitating more effective manure N use on croplands [48]. However, a number of policy barriers exist in the US, as compared to other countries, to integrated crop and livestock systems [51].

There are other important N management practices that focus on the efficient use of N fertilizer. These include soil testing, chemical plant tissue analysis, use of adaptive in-season N recommendation tools to optimize split N fertilizer applications, predictive N management approaches, and sensor-based N management [34]. These strategies are all being developed and tested at land-grant universities in the US, including within the hotspots identified here in table 3.

Finally, in some hot spots intensive production of fruits and/or vegetables (many of which fall under 'other crops' in our cropland typology) is prominent and likely plays an important role in N dynamics (e.g. Washington, Oregon, California, and Florida) (table 3). N fertilization rates for crops such as oranges, lettuce, tomatoes, and potatoes are typically high, with averages ranging from 165 to 241 kg N ha⁻¹ yr⁻¹ based on USDA NASS data collected during 2002–2016 [52]. For comparison, mean N fertilization rates for corn grain, wheat, and other small grains using the same data source were 151, 74, and 53 kg N ha⁻¹ yr⁻¹, respectively.