Inputs into the agricultural system, which are within the grey box in Fig. 1, are from mined phosphate rocks and atmospheric deposition. We did not include P from the in situ weathering of soil particles because the rate of this process is insignificant compared with the magnitude of other inputs (Liu et al., 2008). Outputs included P emission into the atmosphere from fires and P loss to uncultivated land or bodies of water.

We compiled the flows of P in international trade both from the P embodied in crops and livestock products and in P embodied in fertilizers exchanged between countries. For agricultural commodities, we used FAOSTAT data that provided a matrix of commodities exchanged between countries and converted this data into P fluxes using commodity-specific P content data. For P fertilizers, we used the International Fertilizer Industry Association trade statistics. By convention, a positive trade balance for a country means that it is a net P importer. In addition, P fluxes associated with the international trade of fertilizers, food, feed, and fibre commodities can be associated with local cropland PUE and pasture PUE. We defined the dependency on chemical fertilizer imports (Ffer) as the ratio of the P in imported chemical fertilizers (Pfer-imp) to the P in all chemical fertilizers consumed by a country (Pfer-con). Similarly, we defined the dependency on food imports (Ffood) as the ratio of P in food imports (Pfood-imp) to the P in all food consumed by a country. Furthermore, we defined Ftotal as the ratio of the total P imported (food and fertilizers) to the total P consumed as fertilizers and food in a country. The equations for these calculations are presented in Sects. 2 to 6 of the Supplement.

Annual changes in P stocks in cropland and pasture soils (ΔP) were estimated as the difference between inputs and outputs (i.e. the budget); ΔP>0 indicates net P accumulation in the soil, ΔP<0 indicates a net deficit, and ΔP=0 represents no net change. ΔP calculated in this manner does not reflect the legacy effects from previous management and fertilization practices (Ringeval et al., 2014), but it is a useful metric to identify regions with a P surplus or deficit at any point in time and to compare countries.

Annual soil ΔP values were calculated as the differences between annual inputs and outputs. Details and the equations are presented in Sect. 2 of the Supplement.

Following the method of Sattari et al. (2012), we separated the P inputs to soils (except inputs in seeds) into two pools: (1) a stable P pool, which represents P that is unavailable to plants on an annual basis, such as the P absorbed onto iron and aluminum oxides (20 % of total P inputs, including fertilizers, manure, sludge, and deposition); and (2) a labile P pool that is assumed to be available for plant uptake (80 % of total P inputs). P can be exchanged between the two pools. If inputs of labile P are larger than P removal in crop biomass, we assumed that the surplus labile P gets transferred into the stable P pool at the end of the year. In the opposite case, in which inputs of labile P are lower than P removal, plants can take up P from the stable pool (Sattari et al., 2012). This approach assumes that the P loss by run-off and leaching into bodies of water is from the labile P pool only and that P stored in seeds does not belong to either the stable pool or the labile pool. This approach is simplistic, as more research will be required to allow for more realistic modelling of these two pools and the flows they are involved in.

We defined PUE as the ratio of P in the harvested economic outputs to P in the inputs for the entire agricultural system (the grey area in Fig. 1) or for a given subsystem. PUE indicates how much of the input P is transferred into value-added products. If PUE >1, the input of P is insufficient to sustain the output (harvested P), suggesting a net reduction of the system's P reservoir. For cropland PUE, we defined P in harvested crops as the economic P output of the crops and the sum of phosphate fertilizer, livestock manure, human sewage sludge, and P from atmospheric deposition as the P input. For pasture PUE, harvested P refers to the P consumed by grazing animals and the sum of phosphate fertilizer, livestock manure going to the pasture, and P from atmospheric deposition as the total inputs. For the livestock subsystem, the harvested P output represents the P in livestock products (meat, eggs, and milk), whereas the inputs represent the input into livestock. We also defined the PUE of human food (εfood) as the ratio of the P content in human excreta to the total P input in human food; this represents an inconsistency with our previous definitions, since human excreta currently have no economic value. The equations for all the PUE terms are provided in Sect. 5 of the Supplement.

Uncertainties in each flux originate both from the material flux data and from data on the P concentration in each material considered by our analysis, including crop products, crop residues, livestock, meat, eggs, milk, livestock, and human excreta. Many of the global statistical datasets used in our analysis are not replicated, and no alternative dataset is available for establishing a range of uncertainty values for the different P fluxes. National datasets have usually not been formally analysed to determine their uncertainty, and many of the sources of uncertainty are difficult to trace (e.g. clerical errors, differences between countries in product definitions). Thus, we have only addressed the effect of uncertainties in the P concentration by means of Monte Carlo simulations (3000 iterations) using the range of P concentrations reported in the literature (Table S5).

Globally, both cropland and pasture presented soil P accumulation from 2002 to 2010, with an accumulation of 59.6 and 19.4 TgP, respectively. For croplands, the net P accumulation in the stable P pools amounted to 52.7 Tg P, and the remaining 6.9 TgP accumulated in soil labile pools. For pasture, the accumulation in the stable P pool was 25.0 Tg P, but 5.6 TgP was transferred from the stable P pool to be incorporated by grass in regions where P inputs are lower than grass P uptake.

Those global numbers mask large regional differences (Table 1 and Fig. 4) and there were also differences between cropland and grassland. About 32 % of the global cropland area (in 75 countries) had annual soil P deficits from 2002 to 2007, with a net cropland soil P accumulation of 6.20–7.66 TgPyr-1. This fraction increased to 50 % in 2008 and 2009 but the net cropland soil P accumulation decreased to 4.38 TgPyr-1 in 2008 and 5.39 TgPyr-1 at the time of the global financial crisis as a result of high P prices and the resulting reduction in fertilizer application (Cordell et al., 2009, 2012). However, the fraction of cropland soil P deficits returned close to the decadal mean value in 2010, with a net soil P accumulation of 7.30 TgPyr-1. On average, 48 % of cropland P uptake was supplied by stable P that accumulated in previous years according to the equations in Sect. 3 of the Supplement. Including the United States, France, Russia, Argentina, and Paraguay, 89 countries had labile P inputs into cropland that were lower than crop P removal from 2002 to 2010. However, if we consider stable P inputs, cropland soil still presented a net soil P surplus in the United States during the same period. Compared with cropland, a slightly larger proportion of the total global pasture area had a net annual soil P deficit from 2002 to 2010, mostly in Europe and North America. The deficit proportion of grassland was only about 38 % in 2002 and 2003, with an annual net soil P accumulation of 2.26 TgPyr-1; however, it increased to 43 % during the period of 2004–2010 and the annual pasture soil P accumulation was about 2.10 TgPyr-1, with the smallest of 2.00 TgPyr-1 in 2009. However, only 48 countries had labile P inputs into pasture that were lower than the P removal in grass.

Table 1 gives the values of PUE for cropland, pasture, livestock, and food (human use) in the world's different regions. Globally, 116 countries have cropland PUE values above the global mean value of 0.46, mostly in Africa, and these countries account for 64 % of the global cropland area. In addition, 16 % of the countries had a PUE of around 0.6 (0.55 to 0.65). In particular, African countries had the highest overall cropland PUE (≥0.80) because of their low P input. On the other hand, eastern Asia and Oceania have cropland PUE below the global average. Conversely, pasture had high PUE in Europe (1.25) and North America (0.98) but low values in Africa (≤0.77) and particularly low values in the Caribbean and Central America (0.37). P removal from pasture exceeded P inputs in Europe, resulting in pasture PUE >1, largely because of P inputs from feed given to animals.

The livestock subsystem generally had a low PUE (<0.1), with the highest values in Europe, North America, and eastern Asia (Table 1). Regarding human food PUE, our data indicate that only 25 to 40 % of the P in food products in eastern Asia, Oceania, Europe, and North America is actually consumed by humans (Table 1). The resulting low PUE of human use in these regions results from both large P inputs and high food waste. Eastern and southern Africa, western and central Africa, southern and southeastern Asia, and the Caribbean and Central America had the highest PUE for human use, with more than 60 % of P in food being consumed by humans. Globally, most of the P consumed by humans (78 %) originates from crops, and the fraction of P from livestock differs among regions; it ranges from 35 % of the total human food P consumption in Oceania, Europe, and North America to 10 % in less developed regions (Africa and the Caribbean and Central America) and to 4 % in western and central Africa.

Approximately 2.1 TgPyr-1 entered into international trade in 2010, amounting to about 17 % of the total harvested crop P (Fig. 5). The remainder (10.6 TgPyr-1) is consumed domestically. Differences in crop types as a result of their specific P content (Table S1) strongly determine the magnitude of the traded P fluxes. For example, 37 % of the P in soybeans and 27 % of the P in wheat produced each year were traded internationally in 2010. Also significant fractions of the P in maize, other cereals, and fruit were traded internationally, but almost all of the P in sugar crops and fibre were consumed or processed in the countries where they were grown.

Considering the P fluxes in phosphate fertilizers and food products, we examined how international trade influences regional P budgets and redistributes P between regions. We found that southern and southeastern Asia have the largest net P imports (Table 1), with imports of phosphate fertilizer amounting to 1.4 TgPyr-1 and P exports as food products being much smaller, mainly to China and South Korea. South America is the second-largest exporter of P in food, but imports 56 % of its P fertilizer. North America is a large exporter of P in both crop products and fertilizer, yet it also imports P-rich milk products. Most European countries imported nearly all their phosphate fertilizers, but Europe as a whole is a net exporter because of large exports (0.9 TgPyr-1) from Russia (Fig. 6). Western European countries were the main exporters of P-rich livestock products. Some northern African countries (especially Morocco and Tunisia, which have the largest mines of P-rich ores) exported a total of 0.7 TgPyr-1 in fertilizer. The remaining regions (eastern and southern Africa, northern Africa, and the Caribbean and Central America) imported P in both food and fertilizer, although much less than other regions (Table 1).

Figure 6 illustrates the disparities among countries with respect to the role of international trade in crops, livestock, and fertilizer for the main exporters and importers. Based on data for all 224 countries, a country can be categorized into one of the following four groups (Fig. 7).

Food and fertilizer P exporters. P storage in these countries has been decreasing due to their international exports of both fertilizer and food. Examples include the United States and Russia.

Food P importers and fertilizer P exporters. This group mainly comprises countries that export phosphate fertilizers and import food to meet domestic consumption. Examples includes Tunisia, Morocco, and China.

Food P exporters and fertilizer P importers. These countries have high food and livestock production, but this depends strongly on phosphate fertilizer imported from other countries. Examples include Brazil, Argentina, Canada, France, Australia, and India.

Food and fertilizer P importers. These countries depend on imports for both food and fertilizers; they are thus vulnerable to economic shocks that result from changing food prices. Examples include Japan and Indonesia.

International trade affects the global P cycle by physically moving the P contained in traded crops, livestock products, and phosphate fertilizers (Grote et al., 2005). Imports of P fertilizers accounted for 55 and 79 %, respectively, of the total application of P fertilizer for countries that are food P exporters and fertilizer P importers or food and fertilizer P importers. The P trade in food followed a similar trend. Countries that are food P importers and fertilizer P exporters or food and fertilizer P importers depended more on food imports than countries that are food and fertilizer P exporters or food P exporters and fertilizer P importers. International trade also increased the connections among countries (Table 2). For example, although the United States and China are clearly major P fertilizer exporters, they also import fertilizer from each other; 2.6 % of the P fertilizer applied in the United States originated in China, and 3.6 % of the phosphate fertilizer applied in China originated in the United States. In addition, 11.4 % of the phosphate fertilizer consumption in the United States originated from Russia, Morocco, Tunisia, and other countries. About 1.5 % of Chinese domestic P consumption originates from the United States, which is higher than the fraction of domestic P consumption in the United States from China. Countries with small or no reserves of P-containing minerals imported large amounts of phosphate fertilizer; for example, imports accounted for 61 and 46 % of total P consumed in France and Brazil (food P exporters and fertilizer P importers) and 76 % of total P consumed in Japan.

We estimated the net cropland soil P balance in 2000 by means of Monte Carlo simulations, as described in Sect. 2.7. We found a net accumulation of 5.8±0.6 TgPyr-1. More detailed calculations suggest that uncertainty in the crop P concentrations contributed ±0.2 TgPyr-1 of the uncertainty in the net cropland soil P balance; this is because of the dominance of calculations using cereals, which have low uncertainty due to the narrow range of reported P concentrations (Antikainen et al., 2005; COMIFER, 2007; USDA-NRCS, 2009; Waller, 2010). Uncertainty in P concentrations in crop residues contributed an additional ±0.2 TgPyr-1 to the total uncertainty, and uncertainty in P concentrations in the livestock manure applied to cropland added ±0.4 TgPyr-1. In addition, the uncertainty in the pasture soil P balance attributed to uncertainty in the P concentrations in grass biomass and manure was ±1.3 TgPyr-1. This relative uncertainty is higher than that for the cropland soil P balance and the results from the large range of grass P concentrations found in our review of the available data. See Table S5 for more details.

Figure 8a shows the relationship between the cropland PUE and cropland P inputs for 35 countries that are large crop producers. PUE decreased exponentially with increasing input; that is, P was used most efficiently at low application rates. PUE decreased rapidly as P inputs increased to 10 kgPha-1yr-1 and then decreased more slowly. High PUE values were associated with countries that had a low P input and a soil P deficit. This suggests that there is a trade-off between the efficient use of P in cropland and the avoidance of soil P deficits that limit crop yields (Obersteiner et al., 2013). Figure 8a also indicates that cropland soils have a net soil P deficit if their inputs are lower than 10 kgPha-1yr-1, which is a threshold value that corresponds to PUE =0.67. Argentina, South Africa, Indonesia, Mexico, and Paraguay are below this threshold (Fig. 9).

P in harvested crops increased exponentially with increasing P inputs, but the response slowed at high P inputs (Fig. 8b). The P in harvested crops in countries with cropland PUE >0.67 (except Argentina) is only half of that in countries with high P in the harvested crops, such as the United States and China. P in the harvested crops was very low in Australia due to low cropland P input, which was less than 25 % of the inputs in the United States and China. P already present in the soil may be sufficient to sustain high crop yields for some time without additional inputs in some countries (e.g. France) that formerly had large P fertilization rates, despite currently having a negative annual P balance. Comparing Fig. 8a and b suggests that total cropland P inputs of 20 to 25 kgPha-1yr-1 may be a good compromise that will achieve high yields while creating a near-equilibrium soil P balance. Both excessive P inputs (e.g. China and Japan) and low PUE (e.g. India) can lead to high P accumulation in cropland soil, leading to high losses into the environment.

The data in Fig. 8 indicate that different countries face different challenges for P resource management, implying a need for country-specific policy options and solutions. Countries like Kazakhstan and Argentina may have to increase P inputs to their cropland in order to prevent long-term depletion of soil P, which could be realized by increasing the application of phosphate fertilizer or reducing losses to leaching and erosion. Countries like France that are currently experiencing a net negative soil P balance (Fig. 9) following a period of sustained accumulation (Senthilkumar et al., 2012; van Dijk et al., 2016) may need to progressively adjust fertilizer inputs in coming years to balance inputs with removals and avoid the risk of a long-term soil fertility decline due to inadequate levels of P. In contrast, countries such as Japan and China are rapidly accumulating P in cropland soils due to high and sustained P inputs and will urgently need to consider how to improve their cropland PUE. This could be initiated by identifying crop types that are being over-fertilized and regions with excessive application of phosphate fertilizer; they can then consider a range of options such as precision agriculture (i.e. applying only as much P as the crop requires). We estimate that if Chinese cropland PUE could be increased to the global average of 0.46 (Fig. 9), China would save 3.8 TgPyr-1 of phosphate fertilizer, which is equivalent to 60 % of its phosphate fertilizer consumption in 2010. Last, in countries like India where crop P harvests are lower than average despite high average P inputs and positive soil ΔP, improvements in agricultural management (such as the use of precision fertilization) appear necessary. We did not have access to subnational data for this study, but it is likely that in a country as large as India, excessive or insufficient P may occur in different regions, for different crop types, or for different region–crop-type combinations.

Figure 10 shows that the soil P balance is negatively related to the flux of P in livestock products per unit area of pasture. Several western European countries (Germany, the Netherlands, Denmark, and Belgium) achieve high P yields in livestock products (defined by the amount of P in livestock products per unit area of pasture), and all of these countries export livestock products. In these countries, only a small fraction of livestock manure is recycled to pasture, so there is currently a soil P deficit; in the long term, this may result in a loss of soil fertility. Therefore, these countries should increase P fertilization in pasture or import forage or feed to supply the P required to sustain high livestock production. New Zealand, Australia, and Canada are also large exporters of P in livestock products. However, given their low-input production systems and large areas of pasture (Fig. 4b), P removals per unit area through grazing are much lower than in western Europe, and the soil P balance of pasture ranges from slightly negative to slightly positive.

The increasing consumption of livestock products by humans is an essential factor that is responsible for increasing P mining and increasing P inputs to agricultural systems (Metson et al., 2012; van Dijk et al., 2016). Where socioeconomic development is improving the income of residents, especially in Africa and the Caribbean and Central American region, residents are consuming more P from livestock products (Fig. S3). Unfortunately, the livestock PUE in countries in these two regions is much smaller (0.01 to 0.03) than the global average of 0.06 (Table 1), indicating that only a small proportion of livestock P inputs is used by humans. This may be because countries in these regions are primarily importers of livestock products. Therefore, animal husbandry has important implications for global P security and special attention will be required to improve livestock PUE (Wu et al., 2014). If livestock PUE reaches the global level of 0.06 in these two regions, both regions could more than double their livestock production by about 0.16 TgPyr-1.

In addition, the management of manure differs greatly among regions due to different livestock production systems. The yield of livestock products is very low in African countries, resulting in low livestock PUE. Almost all livestock manure is applied to cropland for which this resource is an important P input. In contrast, with the application of phosphate fertilizer to pasture in Europe and eastern Asia, only a small fraction of livestock manure is recycled for pasture (36 and 17 %, respectively); a larger fraction of the manure is applied to cropland in eastern Asia (40 %) and Europe (60 %). Consequently, improving the manure utilization efficiency and applying more livestock manure to pasture will be important strategies in eastern Asia and Europe (Wu et al., 2014).

As shown in Sect. 3.3, only 45 % of the P that enters the food production subsystem was absorbed by humans; thus, large amounts of food (and the P it contains) are wasted, although some parts of the waste were consumed by livestock. Despite this recycling, 2.2 TgPyr-1 flowed into the environment as waste either before or after food consumption, and only 14.3 % of the total P inputs to the agriculture system ended up in food consumed by humans. In eastern Asia, Oceania, Europe, and North America, the PUE of human food was very low, reflecting the high proportion of livestock products in the diet and a high degree of waste. Therefore, decreasing food waste before consumption, recycling P in food waste, and better treatment of organic waste could significantly decrease the amount of P required to support humans (Metson et al., 2012; van Dijk et al., 2016). In eastern Asia, Oceania, Europe, and North America, fully absorbing the 45 % of the P that enters food produced for humans could reduce agricultural inputs of P by 0.7 TgPyr-1 globally. Thus, decreasing food waste and improving the PUE of human food represent key challenges that must be solved to achieve sustainable P management.

Population increases and dietary changes are requiring higher P inputs in cultivated land and the increased mining of P ores (Grote et al., 2005; Foley et al., 2011). From 2002 to 2010, this mining increased by 33 % in our estimate, during a period when the global population and per capita food P consumption increased by 10 and 5 %, respectively. In 2010, humans consumed 8.0 and 3.8 % more P in livestock products and crops, respectively. Since livestock PUE was much lower than cropland PUE, the consumption of more livestock products resulted in lower external P inputs in food that flowed into the human subsystem; this proportion decreased from 36 % in 2002 to 31 % in 2010. Therefore, consuming more livestock products will require increasing P inputs. Thus, human dietary shifts may have been responsible for half of the increase of P-ore mining.

International trade also increased the connections among countries. Whether international trade is good or bad for humans and the environment in terms of its impact on the management of P resources is a complex question. International trade can increase cropland P deficits if countries that export large amounts of P in crop and livestock products do not counteract these exports by increasing inputs of phosphate fertilizer to soils. For example, Argentina exported lots of food to other countries (about 0.15 TgPyr-1) and has developed a serious cropland soil P deficit of 0.38 TgPyr-1 (10.3 kgPha-1yr-1). Massive P imports through trade can result in an excess supply of P to cropland soils as manure (Schipanski and Bennett, 2012), with potentially significant negative environmental effects. On the one hand, trade can hamper the proper recycling of P resources from waste and manure to agricultural soils through local food webs (Schipanski and Bennett, 2012). On the other hand, trade may contribute to more efficient use of P resources if traded products flow from countries with lower PUE to countries with higher PUE, as is generally observed for water resources (Dalin et al., 2014). This confirms that more integrated studies are required to fully assess the effects of trade on P resource recycling, efficiency, and conservation. Our study identified world regions and countries with lower PUE and others with high PUE and regions and countries with a net loss of P in soils and others with a net gain. This provides valuable information to policymakers on how to improve the trade relationships for a global optimization of PUE and therefore global food security.

Previous studies have estimated P flows in agriculture on a global scale (Smil, 2000; Sheldrick et al., 2003; Liu et al., 2008; Cordell et al., 2009; Bouwman et al., 2009, 2013; Potter et al., 2010; MacDonald et al., 2011). However, to the best of our knowledge, the present analysis provides the first consistent multi-year overview of the P flows in agriculture. In addition, it provides national and regional P budgets, calculates agricultural PUE, and quantifies P fluxes in international trade based on a combination of datasets for cropland and pasture inputs (fertilizers, manure, atmospheric deposition, and recycling of crop residues) and outputs (crop harvests, residue removal, and P loss by burning and leaching or surface run-off into bodies of water). For data from 2000, our results are consistent with the abovementioned studies for most P flows (Table 3). For data from 2000, our results are generally consistent with those in the previous studies for cropland soil P inputs, harvested crop P, cropland soil P lost by erosion or surface run-off into bodies of water, pasture soil P inputs, and harvested grass P (Table 3). However, methods, data sources, and system boundaries differed among the studies, making an accurate comparison difficult. Our estimate of a net accumulation of 5.8±0.6 TgPyr-1 is in line with the reported net accumulation in soils, which ranged between 0 and 11.5 TgPyr-1 (Smil, 2000; Bennett et al., 2001; Bouwman et al., 2009; MacDonald et al., 2011), but disagrees with the estimate of Liu et al. (2008), who calculated a net loss of 9.6 TgPyr-1. The difference from the present results can be explained by accounting for large P losses (19.3 TgPyr-1) due to soil erosion caused by land-use change and over-grazing. The quantification of erosional losses of P from arable land is prone to high uncertainties due to the unknown amount of redeposited soil material, and other studies have reported much lower losses (e.g. 2.5 TgPyr-1; Quinton et al., 2010).

The main cropland P fluxes estimated in our study agreed with previous results, except for the production and recycling of crop residues (Table 3). Smil (2000) and Liu et al. (2008) used harvest index data (defined as the ratio of total aboveground biomass to crop residues) for estimating the P in crop residues, whereas we estimated P in crop residues by combining data from Liu et al. (2008) and FAO. MacDonald et al. (2011) estimated that 29 % of the global cropland area was subject to soil P deficits in 2000, which is similar to our estimate (32 %) based on data from 2002 to 2010. In addition, our estimate of 22.3 TgPyr-1 in animal manure for the livestock subsystem in 2000 is within the reported range of 17.1 to 24.3 TgPyr-1 from Potter et al. (2010). We defined global cropland PUE as the ratio of P in harvested crops to total P inputs without accounting for the recycling of crop residues. Under this definition, global PUE was estimated to be 0.43 by Liu et al. (2008) and 0.40 by Smil (2000), both of which are comparable to our estimate of 0.46 from 2002 to 2010. Since we applied the same methods across the globe to calculate agricultural P fluxes, we were able to compare the P fluxes and budgets for different regions and countries on a consistent basis. This information is of critical importance for the development of more appropriate agricultural policy and to support the development of technological and other solutions for different types of countries, which better integrate cultivated ecosystems, livestock production, and the human food supply.

Due to limited data sources for some parameters, our study and most previous studies focused on P in livestock products and manure as the outputs of the livestock system and did not consider the fate of P in non-edible livestock products (e.g. bones, blood, leather products). Xu et al. (2005) pointed out that from 12 to 23 and 72 % of P were contained in livestock meat and bones, respectively. If these percentages are applied to our data, this gives an annual flux of 2.5 TgPyr-1 in the bones of slaughtered animals. Although most livestock bones are currently wasted or landfilled, some countries have begun to use them as fertilizers, protein sources, and condiments (Wu and Ma, 2005; Li, 2008). In addition, as we focused on the annual P budgets for livestock and human beings, we did not account for P accumulation in humans. From 2002 to 2010, the global population increased by 635×106 persons. If we assume that a typical adult body contains 600 g of P, then about 0.38 Tg more P would have accumulated in humans. Therefore, the annual human P accumulation would be 0.04 TgPyr-1, accounting for only 0.3 % of the P inputs into humans.

Despite the abovementioned limitations in our study, we were able to achieve some interesting and novel results. First, we have provided a detailed and harmonized summary of the P fluxes as inputs and outputs for the agricultural system and the internal P flows within the agricultural system on national, regional, and global scales. In addition, we have characterized the P budgets and P-use efficiencies in the subsystems of the overall agricultural system and discussed their influences and impacts. Finally, we have discussed how changes in population, diets, and food consumption have influenced the global mining of P ore and how international trade has influenced P fluxes. These insights will support the development of policies to use P more sustainably at national, regional, and global levels.