The Impact of Industrial Activity on the Amount of Atmospheric O2

2.1 Production Chemistry

Iron and steel are produced primarily from the important iron-bearing minerals hematite ${(Fe}_2{O}_3)$ and magnetite ${(Fe}_3{O}_4)$.

At present, the great majority of hematite is turned into iron by reacting the mineral with carbon monoxide in the presence of heat. This strips the oxygen off the iron and turns the CO into CO₂. Although this process typically occurs in three steps, the net reaction is

The carbon in this process typically arrives as coke (derived from coal) which is then partially combusted to form CO.

As written, this reaction shows that 3 moles of O₂ will be required for every 4 moles of iron produced. However, the central question is where the O₂ that ends up in CO₂ comes from. Of the 12 oxygen atoms released in CO₂, 6 of them came from the hematite, effectively yielding a source of 3 moles of O₂ to the atmosphere.

The “O₂ source” nature of the process becomes clearer if we recogize that the F_ff formulation of Eq. 1 assumes the oxidant for fossil fuels is exclusively atmospheric O₂. If there is a non-atmospheric oxidant, the equation needs a correction term. In particular, in the reductive refining of metals, the ore itself oxidizes the carbon in the fossil fuel, yielding reduced metal and CO₂ as the products with little or no atmospheric O₂ required. Thus, Eq. 1 overestimates O₂ loss (relative to CO₂ production) whenever oxidized metals are refined using fossil fuel carbon.

As a specific example, consider only the hematite reduction:

This immediately shows that producing 4 moles of reduced Fe will release 3 moles of O₂. No matter what the method of hematite processing, because the end product is fully reduced iron, the Fe:O₂ ratio remains 4:3.

Magnetite is processed slightly differently, with Fe₃O₄ and CO reacting to yield FeO and CO₂. The FeO is further reduced with CO to yield pure iron with the following net reaction:

Not surprisingly, the Fe:O₂ ratio is 3:2.

2.2 Inventory

We use data from the US Geological Survey Mineral Yearbooks (Tuck, 2020a, b) to quantify the global annual production of iron ore from 1990 through 2018. Values are given in Table 1, including both hematite and magnetite. We convert mass of ore to moles of hematite and magnetite by assuming that 80% of the ore is hematite (Tuck, 2019), but as shown below our results are relatively insensitive to this assumption. Using the Fe:O₂ molar ratios of 4:3 and 3:2 respectively yields the source of atmospheric O₂ associated with pig iron production presented in Table 1. A time series of these values is shown in Figure 1.

Figure 1

The details of the fate of iron after it is mined are complicated, with various steps of concentration and refining yielding pig iron and ultimately steel. There are losses along the way, producing substantial quantities of slag and other byproducts. Nonetheless, effectively 100% of the oxidized iron extracted from the earth is reduced (Tuck, 2021), enabling the calculation of the atmospheric O₂ source given above.

In practice, pig iron contains small amounts of carbon, sulfur, silicon, phosphorus and manganese; impurities that are removed as the pig iron is refined into steel. The dominant impurity is carbon, but since it enters the iron from coking coal during smelting its oxidation has been correctly captured by F_ff in Eq. 1. Sulfur impurities are also almost entirely introduced by the coking coal, so their oxidation will be described by the treatment of sulfur in fossil fuels (see Sec. 5). Silicon is present in the ore in an oxidized state and remains so during smelting (SiO₂ + CaO → CaSiO₃). Phosporus and manganese are more complicated, as they are present in mineral ore in an oxidized state, are reduced early in processing, and are then re-oxidized at a later step. Thus, just as with silicon, phosphorus and manganese have no net O₂ flux associated with processing.

We connect these production data to Z_metals by the Fe:O₂ ratios given above. For example, if the world produces P moles of reduced iron in a time t and h% of that iron comes from hematite ore, the effective flux of O₂ to the atmosphere will be $Z_{{metals}_{hematite}}=\frac{3}{4}\frac{hP}{t{M}_{Fe}}$ where M_Fe is the molar mass of iron.

3.1 Production Chemistry

Essentially all aluminium extracted for human use comes from bauxite, an ore containing various hydrated aluminium oxides, along with iron and other impurities. The primary oxides are diaspore and böhmite (both AlO(OH)), and gibbsite (Al(OH)₃). About 85% of bauxite that is mined is converted to alumina (Al₂O₃) (Merrill, 2022) using the Bayer process of leaching and calcination. The abundances (both relative and total) of AlO(OH) and Al(OH)₃ vary substantially from one deposit of bauxite to the next, but these two species have the same oxygen yields during refining. The remaining 15% of the bauxite is used for a variety of purposes but the aluminium in it is not reduced (Bray, 2021).

Of the alumina produced, the great majority (88% in 2017) (Bray, 2020a, b) is reduced to pure aluminium using the Hall-Heroult process. The remaining alumina is not further reduced, and is used instead in chemicals, abrasives and other products (Bray, 2021).

While the Bayer process (converting bauxite to alumina) is quite complicated, the net effect is a partial reduction of the hydrated oxides. Most importantly for our purposes, the liberated oxygen ends up as water. Thus, ignoring links between the oxygen and water cycles, the production of alumina has no impact on atmospheric O₂.

In contrast, the reduction of alumina to aluminium using Hall-Heroult involves an electrochemical transfer of oxygen from the alumina to carbon from graphite anodes. The consequent release of CO₂ is, once again, effectively a source of atmospheric O₂ since there is oxidation of carbon with no impact on atmospheric O₂ levels. Since the reduction of alumina can be represented as

the ratio of aluminium to O₂ is 4:3.

3.2 Inventory

As with iron, we use data from the US Geological Survey Mineral Yearbooks (Bray, 2020a, b) for global annual production of pure aluminium from 1990 through 2017. Values are given in Table 2.

Since there is no O₂ flux associated with the Bayer process, we only need to account for the refining of alumina to aluminium. Applying the 4:3 ratio given above yields the values shown in Table 2 as well as Figure 1.

4.1 Production Chemistry

The chemistry of copper production is more complicated than that of iron and aluminium since some copper-bearing minerals are sulfides while others are oxides. The dominant sulfides and their approximate fractional molar abundances are chalcopyrite (CuFeS₂ 70%), chalcocite (Cu₂S 15%) and bornite (Cu₅FeS₄ 10%). The oxides are primarily malachite (Cu₂(CO₃)(OH)₂), atacamite (Cu₂Cl(OH)₃), brochantite (Cu₄SO₄(OH)₆) and chrysocolla ((Cu,Al)₂H₂Si₂O₅(OH)₄· nH₂O), all roughly equally abundant (Sillitoe, 2021).

Sulfides account for about 85% of copper production (Schlesinger et al., 2011). Because the sulfides contain no oxygen, they cannot be a source to the atmosphere. Instead, the reduced sulfur and iron liberated during processing are eventually oxidized, functioning as a sink of atmospheric O₂.

Elemental copper in chalcopyrite and bornite is extracted by concentration and smelting, followed by electrolytic refinement. The iron largely ends up in slag as FeO or Fe₂O₃ and the sulfur is emitted as gaseous sulfur dioxide (SO₂). Copper in chalcocite is refined by leaching, followed by solvent extraction and electrowinning (SX-EW). This yields elemental sulfur. Regardless of the mineral type, the extraction method, or the intermediate oxidation state, the sulfur that is liberated nearly always ends up as sulfuric acid with three of the oxygen atoms in the H₂SO₄ coming from gaseous oxygen while the last comes from liquid water (Schlesinger et al., 2011). In the calculations that follow, as with aluminium we ignore any linkage between atmospheric O₂ and the water cycle. For chalcocite this implies a sink of O₂ in the molar ratio 4:3 (Cu:O₂). For chalcopyrite and bornite, the ratios are 8:29 and 40:53 respectively, assuming equal amounts of ferric and ferrous oxide in the slag.

An additional complication for the sulfides is the risk of double-counting O₂ fluxes associated with the oxidation of sulfur, since we develop a separate O₂ budget for this element. We choose to include the O₂ sink associated with copper sulfides in the copper budget rather than the sulfur budget, allowing us to limit the sulfur budget calculations to the reduced sulfur present in fossil fuels.

For the copper oxides, elemental copper is produced using the same methods employed for chalcocite, described above (leaching and SX-EW). The chemistry of the leaching is complicated by the variety of oxides to be considered and by the aqueous nature of the process. For example, when a single molecule of malachite is leached (by sulfuric acid), its five oxygens end up in one CO₂ and three water molecules. Rather than try to account for the many possiblities, we simply ignore the small source of atmospheric O₂ originating from processing of copper oxides, recognizing that oxides only account for 15% of copper production which is itself a small fraction of iron production.

4.2 Inventory

Again, we use data from the US Geological Survey Mineral Yearbooks (Flanagan, 2021) for global annual production of copper ore from 1990 through 2017. Values are given in Table 3.

We assume 85% sulfides, with the relative mineral species proportions (scaled to 100% abundance) and Cu:O₂ ratios given above. Ignoring any O₂ flux associated with the processing of oxides, we obtain the sink of atmospheric O₂ given in Table 3 and shown in Figure 1. We recognize that ignoring oxides leads to a slightly exaggerated atmospheric O₂ sink and discuss this further in Section 6.

6.1 Metals

Collectively, Tables 1, 2 and 3 show that the production of useful iron, aluminium and copper from mined minerals resulted in a net annual flux of O₂ to the atmosphere of between 21 and 22 Tmol a^–1 in 2017, depending on assumptions about the mix of iron species refined into metals. Averaged over the period 2000–2010, the range is 11.6–12.2 with a most likely value of 12.0 Tmol a^–1. This flux (Z_metals) should be added into Eq. 1. Like all of the other terms in Eqs. 1 and 2, Z_metals varies over time. To facilitate incorporation of Z_metals in future O₂-based carbon budgets, the total flux shown in Figure 1 is also given (with uncertainties) as Table 4. The uncertainty in these fluxes is difficult to rigorously quantify due to the assumptions involved in their estimation.

The largest uncertainty of which we are aware comes from the hematite/magnetite mix from which iron is refined. While an 80:20 (mol mol^–1) mix is our best estimate, we only know with certainty that a majority of iron comes from hematite but some comes from magnetite. Thus, we consider 95:5 and 50:50 mixes as extreme cases around our 80:20 central value. These result in an asymmetric range of values for the estimated O₂ flux from iron. In 2018, the effective oxygen release was $(21.37\frac{+0.34}{-0.68}) Tmol\hspace{0.17em}\hspace{0.17em}{O}_2$.

Luckily, uncertain aspects of the processing of iron (e.g. losses to slag and other byproducts as iron ore becomes pig iron, or losses during transport) are irrelevant, as essentially 100% of the iron in the ore is reduced, regardless of whether it ends up in its intended form (Tuck, 2021).

We take the total production figures for iron, aluminium and copper from the USGS, which does not formally calculate uncertainties. Although presented by convention with three significant figures, the global values for iron production are likely good to 0.75% or better (Tuck, 2022c). Global copper production is significantly less certain (roughly 5% for solvent-extraction/electrowinning and 10% for primary smelting (Flanagan, 2022)). We assume similar uncertainty for aluminium (10%). For 2018, these imply uncertainties in the total effective release of O₂ of 0.15, 0.03 and 0.18 Tmol of O₂, respectively.

Our treatment of copper has several additional uncertainties and biases. We assume that iron present in copper sulfides is equally likely to be found in FeO and Fe₂O₃ states, but the true distribution is unknown. We assume an 85:15 split on copper sulfides and oxides, but do not know the uncertainty in that split. Lastly, we completely ignore the roughly 15% that are oxides because of the diversity of oxide minerals and the complexity of the processing. This omission will bias Z_metals low.

While our ignorance of uncertainties in copper production is unfortunate, it has little impact on the uncertainty in Z_metals simply because production of copper is so much smaller than production of iron. Even if the O₂ fluxes from copper carry a 25% uncertainty (an arbitrary, but very conservative number, allowing for the production uncertainty given above and 23% uncertainty from other sources), the 2018 O₂ consumption due to processing of copper will be known to ± 0.09 Tmol.

Uncertainty in O₂ fluxes from aluminium are driven entirely by uncertainties in production so the value of 0.18 Tmol of O₂ (for 2018) quoted above is the whole story.

We have limited the scope of our study to iron and aluminium oxides and copper sulfides. The oxide ores of copper, manganese, titanium, tin, vanadium and tungsten make negligible contributions to the oxygen budget simply because we refine less of them. However, sulfide ores of zinc, nickel, lead and molybdenum are more widely exploited. In total, the amount of these metals extracted from sulphides is comparable, by mole, to copper sulfides. We also ignore the passive oxidation of iron pyrite (FeS₂) in mine tailings, which may lead to O₂ fluxes comparable to the other sulfides. Our neglect of these various sulphides biases our value of Z_metals high (since they are O₂ sinks). Conveniently, this neglected-sulfides bias is in the opposite sense of the aforementioned neglected-oxides bias, though it is about 40% smaller so cancellation is incomplete.

While the list of quantified uncertainties given above is neither rigorous nor exhaustive, it is dominated by a single term: the uncertainty in the hematite-magnetite mix. In 2018, this accounts for 0.51 of the total uncertainty in Z_metals of ± 0.57 Tmol a^–1. Other years are similar. Furthermore, the uncertainty in Z_metals is small compared to other uncertainties in the O₂ budget equations.

Z_metals should be compared to the various fluxes already in Eq. 1. During the 2000–2010 period, Keeling and Manning (2014) report annual O₂ fluxes of F_ff = (934 ± 56) Tmol a^–1 and Z_ocean = (44 ± 45) Tmol a^–1. The inferred net land flux is F_land = (96 ± 77) Tmol a^–1. Z_metals (11.6–12.2 Tmol a^–1) is much smaller than F_ff and very likely smaller than Z_ocean. It is also substantially smaller than the uncertainty in F_land (arising from uncertainties in F_ff, Z_ocean and observed dnO₂/dt (Keeling and Manning, 2014)).

Nonetheless, including Z_metals in Eq. 1 gives a more accurate model of the O₂ budget and should be done for O₂-budgeting work going forward. Z_metals becomes still more significant when Atmospheric Potential Oxygen (δ(APO) ≡ δ(O₂/N₂) + α_B(y(CO₂)/y(O₂)) + C, where C is an arbitrary scaling constant and y is the dry mole-fraction of the gas) (Stephens et al., 1998; Ishidoya et al., 2022) is the focus, rather than carbon fluxes. Since Z_metals is solely an O₂ flux, it becomes relatively large when other paired fluxes partially cancel.

The APO-based analysis of Keeling and Manning (2014) found ocean and land sinks of (2.72 ± 0.6) and (1.05 ± 0.84) Pg a^–1 of carbon for the years 2000–2010. Adding Z_metals to Eq. 1 and repeating the analysis, the oceanic carbon sink increases by $({0.144}_{-0.005}^{+0.002}) Pg\hspace{0.33em}{a}^{-1}$ and the land sink decreases equivalently. If we take the average of our extreme values and use half the range as the uncertainty, the ocean and land sinks become (2.86 ± 0.60) and (0.92 ± 0.84) Pg a^–1 of carbon.

Z_metals is perhaps most relevant for the secular trends in δ(APO) and the uptake of heat by the ocean that can be inferred from it. As given by Resplandy et al. (2019) in Table S6, the secular trend of δ APO_climate has a measurement uncertainty of 0.54 per meg a^–1 (where “per meg” is analagous to % but includes a multiplier of 10⁶ instead of 10² Keeling and Shertz, 1992)). Assuming the atmosphere contains 1.77 × 10⁸ Tmol of gas, 20.94% of which is O₂, adding 1 Tmol of O₂ to the atmosphere will change the abundance of O₂ (and by extension, APO) by 0.027 per meg. Applying this relationship, by 2014, Z_metals, reached 0.55 per meg a^–1 and has continued to grow, highlighting the importance of Z_metals, in analysis of this type.

Resplandy et al. (2019) also calculated an ocean heat uptake of (1.29 ± 0.79) × 10²² J a^–1 for the years 1991–2015. In this same period, Z_metals has an average value of (0.317 ± 0.002) per meg a^–1, or (0.368± 0.002) × 10²² J a^–1, this reduces the inferred heat uptake to (0.92 ± 0.79) × 10²² J a^–1. The correction is roughly half of the uncertainty in the inferred value, but as a systematic error, it should be applied.

Looking decades or more ahead, our framework for calculating Z_metals will likely need adjustment. Efforts are underway to refine iron-bearing minerals at lower temperatures using hydrogen-rich reductants, releasing some oxygen as H₂O rather than CO₂. Bauxite will inevitably be replaced by other aluminium-bearing oxides and silicates. For copper and other base metals, the proportion of sulfide ores will increase as we mine deeper deposits that have not been weathered.

Moving from production to consumption, the reduced state of iron has a finite lifetime in our oxygen-rich atmosphere, leading to a sink of atmospheric O₂. This is very likely a small effect for two reasons. First, the absolute amount of rusting is small. About 85% of steel materials are recycled well before they are fully oxidized (Tuck, 2022b) and much of the steel in exterior use is either formulated as stainless, or is coated periodically to prevent oxidation. Second, iron production in recent decades has grown rapidly (Table 1) causing a large disequilibrium between production and decay that renders the rust sink negligible. Should iron production stabilize or slow in the future, this correction to Z_metals will be worthy of further attention.

6.2 Sulfur

As mentioned in Sec.5, we have chosen to defer a rigorous calculation of the O₂ sink resulting from sulfur oxidation. Much of this sink is already included in Eq. 1 due to the formulation of α_ff, but there remains a sink of order 2.4 Tmol a^–1 that is currently missing.

This value (approximate as it is) should be compared to the 12 Tmol a^–1 sink represented by Z_metals. If we were to introduce Z_sulfur = 2.4 Tmol a^–1 into Eq. 1, it would decrease the ocean carbon sink for 2000–2010 by 0.03 Pg a^–1 and increase the land carbon sink by the same amount. This is clearly a very small correction to substantially uncertain terms ((2.86 ± 0.60) and (0.92 ± 0.84) Pg a^–1 for ocean and land, respectively).

Given the small size of the correction and the complexity of a rigorous treatment, we choose to leave a full analysis for later, simply acknowledging that even after including Z_metals, all O₂-based ocean carbon sink estimates are biased very slightly high and all O₂-based land carbon sink estimates are biased very slightly low.