The U.S. consumer phosphorus footprint: where do nitrogen and phosphorus diverge?

Fundamental changes to the global nitrogen (N) and phosphorus (P) cycles, largely driven by increased nutrient use and loss along the food chain, have major social and environmental consequences (Steffen et al 2015). Notably, N and P loading are key drivers of surface water eutrophication and associated harmful algal blooms (Carpenter et al 1998, Diaz and Rosenberg 2008). These blooms can negatively impact recreation, transportation, fisheries, human health, and water treatment costs (Anderson et al 2002). As with the far-reaching social costs of the N 'cascade' as reactive N moves through multiple environmental pools (Galloway et al 2003, Billen et al 2013), excess reactive forms of P associated with human activities can also negatively affect multiple ecosystem services that are vital to human well-being (MacDonald et al 2016).

Recent work has highlighted the importance of reducing both N and P inputs to maintain or improve surface water quality (Conley et al 2009, Kanter and Brownlie 2019) as well as minimize harmful atmospheric N emissions and associated deposition (Sobota et al 2015). Others have highlighted the potential to recover N and P from agricultural and urban waste streams in order to recycle nutrients to agricultural lands, thereby offsetting the need for new N and P fertilizers derived from the energy intensive Haber-Bosch process and non-renewable phosphate rock reserves, respectively (Dumas et al 2011, Trimmer and Guest 2018, Powers et al 2019). Many actions to reduce N loading likely also reduce P loading, but specific crops, animal products, and management practices have different effects on downstream relative abundance of N and P in the environment (Jobbágy and Sala 2014, Cease et al 2015, Bouwman et al 2017, Springmann et al 2018, Penuelas et al 2020).

Both the total amount of nutrients delivered to aquatic systems and the N:P ratio of nutrient loading affects eutrophication (Downing and Mccauley 1992, Martiny et al 2014). Given the important role of nutrient stoichiometry (i.e. the ratios of N and P to each other and to other elements) in regulating ecosystem processes and functions (Sterner and Elser 2002), the N:P associated with agricultural inputs, processing losses, and waste warrants further attention. The availability of N and P relative to ecosystem demand could affect both downstream ecosystem response to agriculture as well as the potential to recycle nutrients to meet agricultural fertilizer needs. For example, the Redfield Ratio, a classic stoichiometric ratio in limnology and oceanography based on a constant ratio of carbon, N and P (106:16:1 moles) in marine phytoplankton, can offer insight on nutrient limitation of ecosystem productivity (Klausmeier et al 2004). Additional integrated perspectives on how management and consumption patterns in the food system influence the release of N and P to downstream ecosystems is needed.

At least three types of approaches, or lenses, have been applied to estimate the magnitude of nutrient flows in systems, including potential release to the environment. These include: (1) nutrient balances or 'budgets', (2) Life Cycle Assessment (LCA), and (3) 'footprint' tools. Each of these approaches has strengths and limitations. For example, nutrient budgets aim to quantify major nutrient fluxes in a defined spatiotemporal realm and have been applied to identify major sources and loss pathways of N and P within countries, regions, watersheds, cities, and even a single building (Hong et al 2012, Chowdhury et al 2014, van Dijk et al 2016, Wielemaker et al 2019). Yet such studies do not typically evaluate the total impact of a particular product, economic sector, or individual consumer behavior related to nutrient dynamics. In contrast, the purpose of LCA is to quantify the embodied impact associated with nutrient losses to the environment from production or consumption of a product (Finnveden et al 2009, Hellweg and Milà i Canals 2014). This provides a more complete view of an activity or sector's impact on nutrient cycling, but it is of limited use in providing insight into how individuals might alter their behavior so as to mitigate their impact on the environment. Such insight is where footprint tools are particularly useful.

Nutrient footprint tools (e.g. the N-Calculator, Leach et al 2012) estimate the amount of nutrients embodied in a product or an entity's consumption activities. As such, they provide a different perspective than budgeting or LCA approaches. Other per-capita N footprint analyses include investigations of N dynamics in multiple countries, institutions, and regions (Gu et al 2013, Leach et al 2013, Leip et al 2014, Stevens et al 2014, Galloway et al 2014, Shibata et al 2014, Hutton et al 2017). There have been comparatively few P footprints developed, especially those that mimic the consumer-oriented approach of the N-Calculator. An exception is Li et al (2019), who combine country-level fertilizer, trade, and consumption data to quantify per capita P burdens relative to the biophysical limits of the P planetary boundary (which they call 'P exceedance footprints'). Also, Oita et al (2020) explicitly compare N and P footprints focused on release to the environment to gain insight into nutrient cycling, nutrient impacts, and potential nutrient mitigation strategies in three countries in Asia.

Here, we present a new consumer-oriented P footprint tool for the U.S. based on domestic agricultural production systems and dietary patterns, which follows as closely as possible the methods and data sources for N footprints by Leach et al (2012), with updated values from Cattell Noll et al (2020). We define the P footprint as the amount of P released to the environment associated with the production and consumption of major food groups for one person annually. Our main objective is to estimate how much P is released through the food system for these major food groups in the U.S. from a consumer perspective. We then compare and contrast the N and P footprints for different foods and overall diets. By closely following the N footprint methodology, our results can be used for comparison and assessment of the N:P ratios across different stages of food production and consumption, as well as the N:P of different foods. We address the following questions related to nutrient stoichiometry:

• (a)  
  What is the difference in the magnitude of N and P food footprints? Do different food groups contribute disproportionately to N versus P footprints?
• (b)  
  What are the N:P ratios associated with nutrient release to the environment for different food groups and wastewater?
• (c)  
  How do the N:P ratios of food footprint components compare to known stoichiometric ratios that could influence ecosystem response?

The P footprint approach described here emulates the N footprint approach first described by Leach et al (2012), with updated methods and values (circa 2013–2015⁸) for N from Cattell Noll et al (2020). The N footprint is defined as the N release associated with food and energy consumption, where food consumption is comprised of emissions associated with the food production chain and the treatment of wastewater (including human excreta). Here we focus both N and P footprints only on agricultural production, food processing, food waste, and wastewater components (herein collectively the 'food footprint' shown in equation 1).⁹

$$\begin{align}& {{Food}}\ {{footprint}} = {{release}}\ {{via}}\ {{wastewater}}\\ & \quad + \mathop \sum \nolimits_{{food\ items}}\ {{virtual}}\ {{release}} \times {{food}}\ {{consumed}}\nonumber \end{align}$$

where releases via wastewater are expressed as the N or P released per capita per year (see section 2.2), virtual releases are expressed as release of N or P to the environment per amount of a nutrient consumed for each food item considered in the diet (see section 2.2), and food consumed is the amount of each food item consumed as N or P by one American per year.

Our food N and P footprints center on the average American diet (magnitude and type of different foods consumed) circa the year 2012 as well as the average level of food processing losses, food waste, and wastewater treatment in that period (sensu Leach et al (2012) and Cattell Noll et al (2020)). We consider 17 major crop and animal types in our calculations, with certain crops used to represent broader food categories (table 1, S1, S2, S3 available online at stacks.iop.org/ERL/15/105022/mmedia); for example, the 'beans' category is based on soybean data. Potential nutrient release to the environment associated with crop production (e.g. amount of excess fertilizer nutrient applied per crop after accounting for crop harvested removal) is estimated by using recommended fertilizer nutrient application rates and production data from the major producing states for each crop in the U.S. To account for each state's share of annual production, we calculate a national weighted average for each crop. Recycling of nutrients in the cropping system (e.g. manure nutrients applied to croplands) and in subsequent steps of the food chain (e.g. byproducts of crop processing) is accounted for by iteratively running a model of releases that calculates the recycled ratio of nutrients from each stage that enters as an input to the next stage (see Cattell Noll et al (2020) for details, and table S1 for an equation representation of P that follows how the N food footprint method is presented). Because our focus is on terrestrial agriculture, we do not include a P footprint for fish consumption although there are P losses associated with aquaculture that draw from terrestrial feed sources, such as oilseed crops (Fry et al 2016, Huang et al 2020). Adding seafood is a next step in developing the P footprint, noting that there are many methodological details that must be considered (Oita et al 2016). However, we note that seafood (including fish) accounted for less than 5% of both per capita protein and calorie supply in the U.S. in 2012 (FAOSTAT 2020) and only 3.4% of the total consumption + production food N footprint (Cattell Noll et al 2020).

While the methods for calculating the P footprint is similar to those previously used to calculate the N footprint, there are some important conceptual (figure 1) and methodological differences (section 2.2 and tables 1, S1, and S2). For P, the potential for internal recycling within agricultural systems is considerably higher than for N (Sattari et al 2012, Rowe et al 2016). As a result, P fate and impacts differ considerably from those of N. Notably, whereas N can be transformed and lost from soils as a gas via denitrification (as well as ammonia volatilization and other reactive forms), there is no significant gaseous loss pathway for P.

Zoom In Zoom Out Reset image size

2.1. N footprint approach

2.2. P footprint approach

We follow the same key steps to calculate virtual P factors (VPF) as for VNFs (figure 2, equation (2), table S1), in addition to using the same data sources whenever possible (tables 1 and S2). As with N, we calculate VPFs standardized (i) per unit of P consumed in each food item and then (ii) per unit of food consumed (kg basis), which can then be multiplied by the amount of each food consumed per individual to calculate the total P footprint for an average U.S. consumer. However, in addition to different environmental release pathways, an important distinction between N and P relates to how they are allocated in biological tissues. Since protein is 16% N, knowing the protein content of crops, animal feeds, and human diets facilitates the calculation of N release once waste and loss factors are known. There is no comparable shortcut for P. The following sections explain where changes needed to be made to calculate the VPFs:

$$\begin{equation}\begin{gathered} Virtual\ release\ of\ P\left( {VPF} \right) \hfill \\ \quad = \frac{{\it{P\ release\ at\ field + crop\ waste + processing\ waste + food\ waste + manure\ release + slaughter\ release }}}{{P\ consumed\ in\ food\ item}}. \hfill \\ \end{gathered} \end{equation}$$

Zoom In Zoom Out Reset image size

Note that releases for manure and slaughter are only applicable to animal products, and these releases specifically refer to the 'loss' not recycled to meet crop needs as per Cattell Noll et al (2020). Also, for animal products P release at field refers to releases associated with feed production.

2.2.1. Food consumption

Because P intake cannot be estimated from protein consumption, we use a national database on U.S. food consumption patterns (by weight and calories) that accounts for food loss (USDA ERS 2019) instead of national average protein consumption reported in the FAOSTAT database [used by Leach et al (2012) to estimate the N footprint]. We converted from total food weights to P amounts on a per-crop or per-food item basis, following USDA (2017).

2.2.2. Crop production

For crops, we closely follow the VNF method to calculate crop-specific VPFs. This involved collecting fertilizer P application rate recommendations for particular crops following similar sources (e.g. university extension agencies) as Cattell Noll et al (2020) for N (see tables 1, S1 and S2); we then weighted these P rates by state-specific crop production totals to calculate a national average. By comparing P application to the amount of P that ends up in the harvested crops and crop residues, we estimate excess P potentially released to the environment (as per steps in figure 2). We acknowledge that actual fertilizer use may deviate from state-level recommendations; as a simple test of this, we examined the case of corn, which is by far the largest crop in the U.S. in terms of area, production, and P fertilizer use (Metson et al 2016b). We compared the national P fertilizer application estimated in our model (Σstate-level P fertilizer rate recommendation× state-level corn area planted) for 29 states to the actual fertilizer use for corn nationally in the U.S. in 2014 (USDA ERS 2019) and found very close agreement for this crop (i.e. ∼904 000 tonnes P yr⁻¹ compared to ∼906 000 tonnes P yr⁻¹, respectively). Data on crop-specific fertilizer application rates however are only available for a few select crops which is why we use recommended application rates. We caution that our estimates for more minor crops or crops in different years may deviate further.

2.2.3. Animal production

A key step in estimating the footprint of animal products is to estimate feed inputs required over an animal's key growth stages. To estimate these, we draw from national recommendations for animal feed rations (table 1). As with human food consumption, animal P intake requires some special considerations. Since the P ingested by some monogastric animals is not completely digestible and some animals receive mineral P supplements, the recommended daily intake (RDI) of feed may not accurately reflect annual P ingestion. Sometimes animal P requirements are met by adding mineral supplements to the feed, but here we assume that P requirements are fulfilled by crops only. MacDonald et al (2012) estimate that less than 5% of annual mineral P use in the U.S. goes to feed supplements, comprising up to 12% of total P intake by livestock animals, particularly for poultry. Feed supplement use is mainly due to the un-digestibility of P in the form of phytate by chickens (Ceylan et al 2003). To account for this discrepancy between the amount of feed intake and P recommendations, we therefore converted recommended broiler and layer chicken feed ratios to a total P intake basis (table 1). Although phytate availability can also be a problem for pigs, recommended feed intake was directly reported as total P already (NRC 2012).

2.2.4. Wastewater¹⁰

We assume that all P consumed by humans as food is ultimately released as excreta to the sanitation system, since adults typically retain <1% of P intake in their bodies (Cordell et al 2009). The per capita amount of P released to the environment after factoring in wastewater treatment was modified from Metson et al (2017) to better match Cattell Noll et al (2020). We account for average P excreted and industrial detergent use per capita, U.S. national average sewage connection rates, and the coverage of primary, secondary, and tertiary treatment plants across the country in 2012 and their respective P removal efficiencies. More specifically, the U.S. EPA (2012) Community Watershed Needs Survey indicates the percentage of the U.S. population connected to centralized sewer systems that are covered by either no discharge (7%), advanced (54%), secondary (38%), and less than secondary treatment (2%) facilities. Since estimates of households relying on decentralized options (e.g. septic, cesspools, chemical toilets, simple pit latrines) are uncertain, we assume that 76% of the U.S. population are connected to central systems based on EPA (2012) and the remainder are from decentralized systems. These coverage fractions were then paired with the P removal efficiencies associated with each broad category of centralized wastewater treatment (10% for primary, 30% for secondary, 90% for advanced and 100% for no discharge, Morse et al (1993)) and septic as a representative of for those relying on decentralized systems (50% (Schellenger and Hellweger 2019)). While only 25% of P settles at the bottom of septic tanks (Lusk et al 2017), further retention happens in the drainage field where effluent is treated; retention for on-site septic and cesspool systems has been estimated between 50% and 75%, where, overtime, soil P can leach towards waterways (Schellenger and Hellweger 2019). Overall, we assume that 63% of P produced per capita is retained based on the combination of data described above. The values and approach used here are similar to other U.S. (Hale et al 2015) and global (Van Drecht et al 2009) studies looking at both N and P in wastewater. Still it is important to acknowledge that specific retention values and treatment technologies vary widely across the U.S., depending a on diverse factors including infrastructure types and prevalent industries, which in turn impact nutrient concentrations in waterways (EPA 2012, Skinner and Maupin 2019); decentralized system retention values likely vary even more as management practices and biophysical conditions like soil type, groundwater connectivity, and seasonality can affect P release to the environment (Lusk et al 2017).

2.3. Sensitivity analysis

2.4. N:P ratio of virtual nutrient factors and footprints

We estimate the average U.S. consumer's total annual P footprint to be ∼4.4 kg of P capita⁻¹ yr⁻¹ released to the environment, which is smaller than the food component of the total N footprint of ∼22.4 kg N capita⁻¹ yr⁻¹ (Cattell Noll et al 2020). This disparity between the absolute N and P footprint magnitudes (figure 3(A)) is expected considering that relatively smaller amounts of P inputs are typically required in agricultural system compared to N because of plant stoichiometric needs as well as management practices. In fact, anthropogenic N mobilization has been much higher than P in an absolute and relative terms globally (e.g. due to biological N fixation and atmospheric deposition pathways for N, as well as fertilizer use (Penuelas et al 2020)). The proportion of the N and P footprints attributable to different foods are largely consistent for N and P but meat contributes a greater fraction to the P footprint than to the N footprint. Animal products comprise 78% of the average per capita P footprint and 72% for N. Beef is the single largest contributor to both N and P footprints, although beef does contribute proportionally more to the P footprint. This trend is in line with multiple previous studies that have shown the disproportionate environmental impacts of beef (Poore and Nemecek 2018, Clark et al 2019), largely due to the relatively poor feed conversion efficiency compared to other animal products and relative inefficiency from a food systems perspective compared to directly consuming crops in human diets (Cassidy et al 2013). After beef, the largest difference in proportional contributions to the footprints comes from wastewater. 5% of the annual per capita P footprint can be attributed to surface water release following wastewater treatment (which includes human excreta as well as industrial detergents); in comparison, N removal via wastewater treatment is less efficient than wastewater P removal, contributing 9% of the N footprint.

Zoom In Zoom Out Reset image size

3.1. VPFs and VNFs per unit of food consumed

Overall patterns of nutrient release to the environment per kilogram of food consumed are largely consistent for N and P, with lower nutrient releases for crops than animal products and large nutrient footprints for beef (figures 4 and S1). One kg of beef consumption is associated with ∼112 g P that does not end up in consumed foods. This impact is more than three-fold that of the next largest food type¹¹ (pork, with ∼34 g P released per kg of food consumed). Beef also has a much larger N footprint than other animal products. Although cheese is both P and N intensive (ranking as the fifth and fourth largest for P and N, respectively), milk has a lower P footprint relative to grains (ranking as the eleventh largest). Fruits have the lowest N and P footprints compared to other vegetal foods (fruits < vegetables < beans < grains, for both N and P). Beans have a mean N and P footprint slightly greater than vegetables, but smaller than that of most grains. Animal products generally have larger P footprints than plant products, in part because they have more opportunities for leakage along production chains. This is illustrated by the beef versus wheat comparison (figure 2). Milk is an exception because milk has a relatively low P content and cows produce a lot of milk over their lifetime (which means the P release per kg of product is low). When multiplied out to reflect annual milk consumption (figure 3) in diets it becomes a more substantial part of the total U.S. footprint.

Zoom In Zoom Out Reset image size

Our sensitivity analysis suggests that these overall trends across food products are robust. Confidence intervals around the P release values do not overlap among animal products or among grains, beans, and vegetable and fruits as category averages (figure 4(A)). This suggests that these relative rankings would likely not change due to state-by-state variability in fertilizer N/P application recommendations or changing the FCRs used in calculating VPFs of animal products (figure 4). The exception is milk, which overlaps with the confidence interval for vegetables.

3.2. VPFs and VNFs per unit of nutrient consumed

3.3. Comparison of P footprint results with other studies

3.4. N:P ratio of the food footprint and its components

For every kg of P released to the environment for food, the average U.S. consumer releases ∼5.2 kg of N, but there is a large amount of variability in the ratio among food groups, and even more so between food consumption and wastewater (figure 5). Because beef and other meats result in so much more N and P release per unit food consumed, at first glance it can seem like other foods are similar to one another and close to the Redfield Ratio (figure 5(A)); however our results indicate that grains and pork have similar N:P release ratios (figure 5(B)). While fruits and milk have N:P ratios of 10.4 and 8.8, respectively, all other foods are between 3.5 (eggs) and 6.5 (beans), meaning that they release relatively more P than N compared to relevant ecological and heath values. In other words, the cumulative release associated with these food items are lower than what phytoplankton require to grow (Redfield ratio), what the average human does consume (human consumption), and what they should consume (RDI) in the U.S. (see figure 5(B)). The human consumption N:P ratio may be below the RDI because even though American's consume more protein (and thus N) than is needed, they proportionally eat more P than is recommended (Cease et al 2015).

Zoom In Zoom Out Reset image size

Interestingly, the ratios associated with human waste are highly variable (seen as difference between wastewater and septic dotted lines in figure 5(B)) and do not closely match that of any food group or recommended dietary consumption. We estimate that wastewater release to the environment has a higher N:P ratio than septic waste, given the mix of centralized systems and septic systems across the U.S.. It could be that N entering waste treatment plants is nitrified and then released to surface waters and requires tertiary treatment to remove nitrate (Holmes et al 2019). In contrast, the septic environment is designed to maximize denitrification, and thus may have a lower N:P ratio than the sewage (Wilhelm et al 1994); although both N and P losses from septic towards waterways can be high (Iverson et al 2018).

As is the case with different food groups overall (figure 5), the relative release of N and P at each step of the production chain varies for individual foods (figure 6). For every 100 units of new N and P there is relatively more release of P from excess fertilization (P > N for fertilizer and manure recycling), post-harvest crop losses have relatively more N than P release for both wheat and beef. The larger P releases related to manure and slaughter processing in beef are expected because most P is in manure and bones instead of in the meat, whereas more N is found in meat because of its protein content. This also explains why losses of N due to food waste and consumption of beef are relatively greater for N than for P in comparison to wheat (i.e. food waste and consumption have similar impacts on N and P footprints). Although many release pathways can be opportunities for recycling, animal manure and carcasses are particularly important potential P recycling pathways (figure 1).

Zoom In Zoom Out Reset image size

In summary, the average U.S. consumer's diet is less efficient in its use of P than N along the full production-to-waste chain (prior to wastewater treatment), and this is in large part because of the releases associated with animal products (except milk, figure 5). However, the decoupling of N from P due to sewage and septic treatment means that the average U.S. consumer releases relatively more N than P to waterways relative to what they consume (figures 5 and 6). If the U.S. replaces septic with connections to central sewer systems it is possible that the discrepancy between N and P removal becomes even more accentuated. For instance more effective removal of P than N from wastewater treatment plants over time has been observed in China, where urban lakes have experienced increasing TN:TP ratios as cities 'improved' to municipal wastewater treatment (Tong et al 2020). The specific values of nutrient ratios (wastewater or other), and the ecological implications of any of these stoichiometric relations likely varies widely across the country. Similarly, the relative release of N and P at different steps along the production chain (figure 6) for different food items will both affect recycling potential and have ecological impacts.

3.5. Implications for management

Environmental footprints offer an opportunity for consumers to reflect on how their actions are linked to nutrient use nationally and its potential effects on downstream ecosystems, and as such, they can be an important goal setting tool (Galli 2015, Leach et al 2016). In this work, we presented a new P footprint for the U.S. that could be matched with an existing N footprint and explored how releases of both nutrients (N:P) varied among food types and along the food production and consumption chain. We did not observe large differences in terms of ranking of food types contributing to the N and P footprints so it is likely that interventions for one nutrient will have beneficial effects on the cycling of the other. Still, there are some interventions that could create different outcomes for N than P (especially with regard to recycling technologies) so it is important that future work continue to systematically quantify N and P flows in order to be able to identify where stoichiometric ratios might cause issues. Our results show relatively large differences between N and P across components of the food system and waste management, suggesting that research on coupled nutrients and N:P ratios would help design beneficial interventions for food production and environmental quality.

We thank Rebecca Clarke and Ashley Tseng for assistance with data compilation. Tim Moore, Elizabeth Castner, and Jennifer Andrews made valuable contributions to this work at an initial workshop on the topic of multi-resource footprint tools. Funding to GSM was provided by the National Academies of Science Research Associateship Program funded through US EPA (Assistance agreement number: CQ-83557701), as well as in-kind time and resources at Linköping University. Funding to GKM was provided by the Trottier Institute for Science and Public Policy (TISPP) and the Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grants program. Funding to JAH was provided by an NSF INFEWS grant (NSF EAR1639458). The views expressed in this article are those of the authors and do not necessarily represent the views or policies of the U.S. Environmental Protection Agency.

All data that support the findings of this study are included within the article (and any supplementary information files).