Photocatalytic chlorine atom production on mineral dust–sea spray aerosols over the North Atlantic

Mak et al. (25) saw the largest ¹³C depletions in CO ever recorded, in air samples taken on Barbados (13.2 °N, 59.5 °W) from 1996 to 1999 (Fig. 1). The Barbados samples were depleted in ¹³C by up to −3‰ relative to the other northern hemisphere locations in episodes occurring from July to December, reducing the seasonal minimum δ¹³C-CO at Barbados to −34‰. For other locations, for example, Tenerife (28.3 °N, 16.5 °W), the seasonal minimum of δ¹³C-CO remained above −31‰ (Fig. 1). As seen in the fig., the seasonal maximum at Barbados was similar to that seen at the other locations, around −26‰. Thus, the depletion events are seasonal, episodic, and location specific.

Mak et al. (25) suggest the observed changes in ¹³CO and C¹⁸O are evidence of the reaction of Cl with CH₄. CO is the first stable product in atmospheric methane oxidation, and δ¹³C-CO is sensitive to even small CH₄ removal by Cl because atmospheric (CO) is more than an order of magnitude lower than [CH₄] (8) and because of the large KIE of the Cl reaction. They estimated that the data could be explained if 3 ppbv of CH₄ were destroyed by reaction with [Cl] = 5 × 10⁴ atoms cm⁻³ over 3 d. However, the known Cl sources could not explain the seasonal depletion in ¹³CO (25).

We incorporated the MDSA mechanism into the global 3-D chemistry-climate model Community Atmosphere Model with Chemistry (CAM-Chem) (Methods) (v4) to investigate whether Cl generated by MDSA photocatalysis could be responsible for the CO isotope signal observed by Mak et al. (25). CAM-Chem is included in the well-documented CESM (Community Earth System Model) framework (26, 27) and includes an updated scheme for halogen chemistry (Cl, Br, and I) (13). The parameters used in the model simulation are based on laboratory and field observations from the North Atlantic region near Barbados, which makes it useful for describing the CO isotope anomaly at Barbados, although interpretation in other regions may be less reliable.

Output from our MDSA model shows good agreement with the timing, episodic variability, and magnitude of δ¹³C-CO observations of Mak (Fig. 1). We will discuss this in detail after we first introduce the mechanism.

MDSA is formed when iron-bearing mineral dust and sea spray aerosols mix (28). Iron is released from the minerals by weathering, and according to the Pourbaix diagram (29), iron oxides and hydroxides, abundant in the crust and in mineral dust, are converted into Fe(III) ions under the oxidizing and acidic conditions found in well-mixed marine and dust aerosols found in the North Atlantic (30). Acidity favors formation of Fe(III) chlorides (FeCl_n^3-n with n = 1.4) by keeping hydroxide concentrations low (31). Fe(III) chlorides have an intense ligand to metal electron transfer absorption activated by sunlight, yielding Fe(II) and oxidized chlorine Cl (14). The Cl atoms can escape directly from the aerosol to the gas phase but most likely combine to escape as Cl₂, which is then quickly photolyzed by sunlight (31). Fe(II) is reoxidized to Fe(III) by O₃, H₂O₂, and other species, making the process catalytic in iron by reforming the iron chloride species (29). When Cl reacts with methane, it forms HCl, which can be reabsorbed by the MDSA, making the process catalytic in Cl as well (see SI Appendix, Fig. S1 for a scheme of the mechanism). While this mechanism resembles the release of iodine from dust reported by Koening et al. (32), in this case, the halogen source is the large reservoir of chloride from sea spray already present in the aerosol phase in contrast to deposition of iodine-containing atmospheric gases on the mineral surface. This critical difference means that MDSA–chlorine is produced close to the surface in the MBL from abundant sea spray aerosol. Using a smog chamber, Wittmer et al. (31) demonstrated that formation of Cl atoms by photooxidation of chloride occurs in aerosols that form when iron oxide is mixed with sea salt under acidic conditions. Until now, this mechanism has not been observed in situ and its global impact was unknown, although the necessary elements are found in the troposphere: Photoactive iron is known to exist above the North Atlantic (33–35) and mineral particles in the MBL accumulate oxidation products and salts (including sea salt) on their surface within a day (28).

Zhu et al. (34) collected mineral aerosol at Barbados and measured the daily variability of Fe(II). They saw a clear diurnal pattern in the concentration of soluble Fe(II), with daytime values (mean 3.7 ng m⁻³) about twice nighttime values (1.5 ng m⁻³), showing that photochemical processing of Fe takes place. The same researchers found that a photochemical steady state is established in which Fe(III) is photoreduced to Fe(II) as quickly as Fe(II) is oxidized to Fe(III) (33), and they showed that there are at most 11.4 Fe(II)-Fe(III) cycles h⁻¹ at maximum solar intensity. This rate is at the low end of the range reported by Wittmer of 6 to 78 h⁻¹ (31); atmospheric conditions could result in lower values, more similar to those assumed here, due to the suppression of this reaction in the presence of environmental sulfate (36). Several mechanisms for the photochemical reduction of Fe(III) are known including the oxidation of organics yielding H₂O₂ (33), but we assume in our model parameters that the dominant photoreduction reaction in MDSA yields OH and/or Cl that both lead to Cl₂ production (31). Considering that these dust particles will be mixed with sea salt and assuming the irradiated combination gives rise to oxidized chlorine via the MDSA mechanism, each cycle would produce one Cl atom. The agenda is to i) determine whether Fe photochemistry in Sahara dust is able to produce Cl atoms, ii) model the MDSA mechanism over the North Atlantic, iii) compare the model prediction with the observed CO isotope anomaly in Barbados, and, if successful, iv) estimate the global impact of MDSA photochemistry.

The MDSA mechanism is implemented in the CESM assuming: i) The cycling rate observed near Barbados between Fe(II) and Fe(III) is fully equated with the rate of Cl production, and ii) the influence of sulfate is reflected in these observations because FeSO₄ (which would reduce the Fe available for Cl production) does not take part in the observed photochemistry cycles (36). We ran the model with and without MDSA chemistry and used a tracer molecule to derive the additional CO generated by the Cl + CH₄ oxidation due to the addition of MDSA. We use this to calculate the change in δ¹³C-CO and add the result to the same inverse model as was used by Mak et al. (25) (see Methods for details). We then evaluate how this affects the predicted seasonal minimum and maximum of the inverse model.

The model results show that MDSA photocatalysis produces 110 ± 10 Cl atoms (active Fe atom)⁻¹ d⁻¹ on average at Barbados, which is 70 g Cl₂ (g active Fe)⁻¹ d⁻¹. The modeled average cycling rate between Fe(II) and Fe(III) of 11 h⁻¹ at noon, is similar to that of the observations, consistent with the implemented reaction rates observed by Zhu (33). According to the model results, the average concentration of photoactive iron in MDSA at Barbados is 2.5 ng m⁻³ with peak concentrations up to 34 ng m⁻³ [similar to observations reported by Zhu (34)], resulting in a Cl₂ production rate of up to 6 × 10⁵ cm⁻³ s⁻¹ (average 2 × 10⁴ cm⁻³ s⁻¹) and peak [Cl] of >10⁵ cm⁻³.

The model output shows good agreement with observation (25) for the location, timing, episodic variability, and magnitude of δ¹³C-CO (Fig. 1). The MDSA model reduced the seasonal minimum of weekly average δ¹³C-CO at Barbados by −2‰ in 1996 and by −4‰ in 1997. Because of the episodic nature of the dust intrusions, the springtime dust does not significantly change the seasonal maximum, changing it by only −0.1‰ and −0.3‰ in 1996 and 1997, respectively. The large episodic variations in δ¹³C-CO at Barbados are also seen in the observations. In our model, at Spitzbergen, both the seasonal minimum and maximum did not change by more than −0.3‰, confirming that the MDSA impact is local and specific to the western North Atlantic. Cl formation in the model depends on specific conditions, including dust concentration, wind speed/sea spray aerosol loading, atmospheric circulation, wind direction, and actinic flux. Mak et al. (25) suggested that chlorine is the cause of the ¹³C anomaly in CO, and our CESM simulation is consistent with our hypothesis that the Cl is associated with Sahara dust episodes.

Our model results show that seasonal Cl₂ production rates up to 6 × 10⁵ cm⁻³ s⁻¹ are needed to explain the CO isotope anomaly at Barbados. Other known Cl production mechanisms are not able to produce a seasonality of this magnitude. For example, acid displacement in sea salt forms gaseous HCl which can then react with OH becoming a large source of Cl above the oceans, Cl (14, 37). We estimate that the maximum Cl production from this source is 3 × 10⁴ cm⁻³ s⁻¹ (Methods), which is more than an order of magnitude too small to explain the observations. In addition, the global seasonality of acid displacement as a Cl source comes from solar-intensity-driven OH and DMS seasonality (6), which makes acid displacement seasonality more apparent away from the equator, and therefore does not match with the uniqueness of the isotope anomaly at Barbados. Finally, nonsea salt sulfate levels (and associated HCl levels) at Barbados reach their seasonal maximum in June/July (38), which is opposite to the observed CO isotope effect. Other substantial Cl sources such as ClO and HOCl have at most a similar order of magnitude and are part of the Cl radical chain with their seasonality depending on an initial input of Cl from sources like OH + HCl. We therefore conclude that MDSA photocatalysis is the necessary mechanism for producing Cl and with the right concentration and spatial and temporal structure to explain the observations.

As was already observed by Mak et al. (25), the observations in Barbados are depleted compared to the observations in Tenerife (Fig. 1). In agreement with this observation, the model predicts much smaller MDSA-related dips in δ¹³C-CO at Tenerife (Fig. 1B) and less frequently. Braunlich (39) discussed the Tenerife ¹³CO record in detail and noted that the summer minimum in Tenerife was lower in 1997 compared to 1996, 1998, and 1999. This was also observed in stations at Spitzbergen and Sonnblick, but no explanation could be found. This period aligns with the prediction by our model of significant Cl production by the MDSA mechanism in Tenerife for August 1997.

The modeled annual average shift in δ¹³C-CO across the North Atlantic region shows a shift of up to −4‰ at 30 °W over the North Atlantic (Fig. 2; monthly in SI Appendix, Fig. S2). The zone of maximum shift changes through the year depending on winds in the Inter-Tropical Convergence Zone that carry Sahara dust to Barbados from June through December. Even though more dust is transported across the Atlantic in summer, the model predicted that the strongest δ¹³C-CO excursions at Barbados would occur in the fall as is observed. This is because in summer, most dust is transported at high altitude, above the MBL, which limits mixing with sea spray aerosol (SI Appendix, Fig. S3). This agrees with satellite observations (40) and in situ measurements (41) that show that the Saharan Air Layer transports dust at a higher altitude and velocity in summer than fall. Depletion in ¹³C-CO is also seen in the spring (Fig. 1). While Sahara dust is farther south of Barbados this time of year, its impact is still strong enough to influence δ¹³C-CO at Barbados. However, because of the episodic nature of the dust intrusions, the springtime dust photocatalysis does not significantly change the seasonal maximum.

MDSA photochemistry increases the monthly average [Cl] to >10⁴ cm⁻³ over the equatorial Eastern Atlantic (Fig. 3A) greatly increasing the fraction of CH₄ oxidized by Cl. Fig. 3B shows the reduction in O₃, which is a precursor for OH and leads to a reduction in OH shown in Fig. 3C. The overall effect on CH₄ loss is the sum of the additional loss via Cl and the reduced loss via OH. For most of the North Atlantic, we find that CH₄ removal increases with the addition of the MDSA mechanism (Fig. 3D and SI Appendix, Fig. S7); the local CH₄ loss rate increases by 20%, even though OH is reduced by up to 10% (Fig. 3C and SI Appendix, Fig. S6).

For the North Atlantic region (0 °N to 35° N; 90° W to 0° W), an annual average burden of 0.22 Gg photochemically active Fe in MDSA in the MBL above the ocean surface resulted in a Cl₂ production of 2.9 Tg y⁻¹ [36 g Cl₂ (g active iron)⁻¹ d⁻¹], resulting in a net CH₄ loss of 0.6 Tg y⁻¹ [7.5 g CH₄ (g active iron)⁻¹ d⁻¹], from an increased Cl-mediated CH₄ loss of 1.1 Tg y⁻¹ and a reduced OH-mediated loss of 0.5 Tg y⁻¹ (SI Appendix, Table S1). The total burden of tropospheric O₃ was reduced by 6% above the North Atlantic (SI Appendix, Table S1).

These model results suggest that MDSA produce a large amount of Cl₂ that is not considered in existing Cl budgets (11): 3.8 Tg Cl₂ y⁻¹ across the entire North Atlantic (continental and marine), of which 2.9 Tg y⁻¹ are over marine areas. Globally, the MDSA Cl source increased the tropospheric inorganic Cl source in our CESM simulations by ~41%, adding 13 Tg y⁻¹ of chlorine to 23 Tg y⁻¹ from sea-salt cycling, 8 Tg y⁻¹ from anthropogenic emission, and 1 Tg y⁻¹ from photochemical oxidation of organic chlorine. Because the MDSA source is very strong over the North Atlantic, this increases the Cl contribution to CH₄ oxidation by at least fourfold, making the MDSA mechanism the dominant source of atomic chlorine in this region. Outside of the North Atlantic region, our model predicted Cl₂ production by MDSA in dust from the east coast of Argentina, as well as in the Gulf of Aden, around India and around Australia.

The model employed in this work is a global chemistry–climate model CAM-Chem, which is included in the well-documented CESM framework (CESM) (26, 27), and incorporates an updated halogen chemistry scheme for chlorine, bromine, and iodine (14). In this work, CAM-Chem was configured with a horizontal resolution of 0.9° latitude by 1.25° longitude and 56 levels from the surface to the stratosphere.

For this implementation of the MDSA, we use a simplified mechanism to describe the overall process based on results from field studies. At present, there is no model that combines iron aerosols and chlorine chemistry to describe the elementary processes in detail. In our parameterization, the amount of MDSA was calculated by estimating the amount of photoactive iron. In studies of Fe(II) and photoreducible Fe(III) in North Atlantic dust aerosols, it was observed that photoreducible Fe(III) is found mainly in fine dust particles of sizes below 2.5 µm and across a wide range of dust loads (46–48). These observations imply that a substantial fraction of these dust/salt aerosols are in the acceptable pH range of the Pourbaix diagram (29) for the reoxidation of Fe(II) to Fe(III). We used these observations as input for our model. We used the size bin of 1 to 2.5 µm from the model (referred to as χ_DST02 below) because at least 90% of Sahara dust <2.5 µm is in the 1 to 2.5 µm range, and χ_DST02 therefore provides us with a good reference value to compare with the observations. Additionally, the smallest size fraction is often associated with high sulfate concentrations (30), which could suppress the MDSA (49). In situ observations suggest that most fine particles downwind of the Saharan plume will be acidic enough for the MDSA mechanism (30, 50).

The following equation was used to calculate the amount Fe in MDSA.

where α = 0.063% is the mass fraction of photoactive Fe in the dust, c is a pressure-dependent factor converting χ from kg kg⁻¹ in the CESM model to kg m⁻³, and γ is a parameter set to 1 or 0 depending on the availability of Cl as described below.

This mechanism should be active only in downwind regions from dust plumes and in the MBL since this will assure a high-enough Cl:Fe ratio, above 13 for MDSA concentrations up to 90 ng m⁻³ (31). In order to ensure these characteristics, we limit the Cl production using the constraints ρSSLT02 > 10⁻⁹ and altitude below 900 hPa, by setting γ = 1, and otherwise setting γ = 0. Mineral particles in the MBL accumulate oxidation products and salts including sea salt on their surface (28), with 80 to 90% of the dust particles becoming mixed with sea salt within a day.

The factor α is calculated by multiplying the fraction of Fe in the dust (3.5%) with the fraction of the iron that is soluble and photoactive (1.8%). We used 3.5% as the amount of iron in the Saharan dust, based on the findings of Trapp et al. 2010 (48), as a typical value for dust-sea salt mixtures across the North Atlantic.

Observational evidence suggests an inverse relationship between dust amount and solubility (51). The two explanations for this behavior are 1) differences in particle size (smaller combustion-sourced particles having higher solubility than dust-sourced particles) and 2) downwind atmospheric processing of particles increasing the solubility and the higher gravitational settling rate of larger particles (52). However, the fraction of the total iron that is photoactive does not appear to follow this inverse relationship at locations downstream from the dust sources (48) because observations across the North Atlantic show a similar fraction of total Fe to be photoactive, for both high and low dust loads (46–48). Therefore, we assume a constant value for the fraction of the iron that is photoactive for the long range–transported dust of 1.8% based on cruises far from North Africa (35, 46–48) and average values for the model-data syntheses for dust in the North Atlantic (52). Our parameter of 1.8% photoactive soluble iron is in line with the findings reported by Zhu et al. (34) who measured the daily variability of Fe(II) at Barbados in September 1992 and reported a clear diurnal pattern in the concentration of soluble Fe(II), with daytime values (mean 3.7 ng m⁻³) about twice nighttime values (1.5 ng m⁻³). This observed cycling of Fe(II) and Fe(III) during the day and night demonstrated that photochemical processing of Fe does take place at Barbados, and we postulate here that it is entirely due to the Cl mechanism. Next, the MDSA number w_MDSA is converted to d_MDSA in mol Fe m⁻³, using M_w(Fe) = 55.845 g mol⁻¹.

Zhu et al. (33) studied Fe(III) photochemistry for typical (aerosol) conditions at Barbados. They found that for constant solar intensity, a steady-state is established in which Fe(III) is photoreduced to Fe(II) as quickly as Fe(II) is oxidized to Fe(III) by H₂O₂. They demonstrated experimentally that in a typical aerosol at Barbados, the oxidation of Fe(II) to Fe(III) by H₂O₂ has a reaction rate constant of 0.19 min⁻¹ at a typical H₂O₂ concentration of 50 µM at pH = 1. They also showed that during most of the day, this balance remained stable at high Fe(II) levels, which indicates that photoreduction is much faster than oxidation. Using the reaction rate of 0.19 min⁻¹ would allow for a maximum of 11.4 photochemical cycles h⁻¹ if photoreduction is not the limiting step. This is in good agreement with the values found by Wittmer (31) of 6 to 78 h⁻¹, although the Fe-Cl reaction is sensitive to chemical composition and is partly suppressed in the presence of sulfate (33).

The MDSA aerosol chemistry is described by Wittmer (36, 49). Fe(III) chlorides FeCl_n^3-n (n = 1.4) have intense UV charge-transfer absorptions. On absorption of light, an electron is transferred from chloride to Fe(III), resulting in photoreduction of Fe(III) to Fe(II) and yielding chlorine atoms (Cl) in the liquid phase. Chlorine atom recombination is catalyzed by Cl⁻ via Cl₂⁻ leading to Cl₂ which degasses.

In summary, Cl production occurs at the rate at which iron is cycled between Fe(II) and Fe(III) within d_MDSA. Assuming steady state, the fraction of d_MDSA that is in the Fe(II) state is given by

In which k₁ is the reaction rate constant for oxidation of Fe(II) to Fe(III), and k₂ is the reaction rate constant for photoreduction of Fe(III) to Fe(II) at solar intensity I₀ = 600 W m⁻², and I is the temporally varying solar intensity.

At steady state, the production of Cl can be calculated from either Fe(II) oxidation or Fe(III) photoreduction. Using Fe(II) oxidation, the Cl production rate (amount per unit time; $r_{\mathit{Cl}}$ ) is

Combining Eqs. 1 to 4 and using two Cl atoms to create one Cl₂ molecule gives the following rate of Cl₂ production:

For oxidation, we used the reaction rate constants from Zhu et al. (30), k₁ = 0.19 min⁻¹. Because field measurements suggest that photoreduction is not the limiting factor, we used a photoreduction reaction rate of 10 times as fast: k₂ = 1.9 min⁻¹. The resulting Cl₂ production rate is in min⁻¹ as well. These values of k₁ and k₂ will result in a maximum of 10.2 photochemistry cycles per hour at maximum sunlight, which is well within the range of cycles measured by Wittmer (31) of 6 to 78 Cl h⁻¹.

To monitor the reaction of Cl + CH₄, we added a tracer to the model with a lifetime of 60 d, representative of the lifetime of CO in the North Atlantic region, i.e., 0° N to 35° N; 90° W to 0° W), based on Pfister et al. (53). We then ran the model with and without MDSA chemistry, using a stabilization time of 12 mo. Using a CO yield of 90%, the difference in the tracer molecule represented the extra CO generated by the Cl + CH₄ oxidation due to MDSA implementation.

To calculate the isotope shift in the CO produced via CH₄ + Cl, we used 100‰ depletion in ¹³C relative to the ambient CO without the MDSA mechanism. The MDSA implementation also changed the amount of CO produced by CH₄ + OH slightly; the isotope shift from this effect was less than 0.05‰ for a 1 ppb change in CO, calculated using the Rayleigh fractionation equation, and was therefore further neglected in the analysis.

Samples were collected weekly at Ragged Point in Barbados in 1996 to 1999, at the top of a 15-m tower, which sits atop a 15-m bluff overlooking the Atlantic. The vast majority of samples were collected during an east-south-east flow in the morning (8:00 to 12:00) with wind speeds in excess of 10 m s⁻¹. CO isotope analysis was done with an isotope ratio mass spectrometer. More details on the sample collection protocol and analysis of CO isotopes are provided in Mak et al. (25). Mak et al. (25) data were shown as 3-wk averages, but in our Fig. 1, we show the raw data that have not been published before.

Fig. 1 shows the same inverse model simulations as used by Mak et al. (25), which is an inverse model simulation using TM2 from Bergamaschi et al. (54). It was noted that these model simulations are for a perpetual year reflecting the climatological wind fields. Model simulations were tuned to a set of previously published stable isotope data which were collected from middle- and high-latitude stations in both hemispheres.

We note that perfect agreement between model and observations is not expected as the Mak et al. (25) samples were taken with nominal conditions of 10 m s⁻¹ wind from the east-south-east, giving an increased chance of sampling air parcels affected by dust, whereas the model data present δ¹³C-CO values predicted for the region regardless of local meteorological conditions.

Using the reaction rate constant for HCl + OH, k(298 K) = 7.86 × 10⁻¹³ cm³ s⁻¹ (55) and a relatively high mole fraction of x_HCl = 400 ppt HCl (7) and a x_OH of 0.14 ppt (double the diurnal average of 0.07 ppt, see SI Appendix, Fig. S6), we calculate a maximum Cl production rate from acid displacement of 3 × 10⁴ cm⁻³ s⁻¹. This is more than 1 order of magnitude below the Cl production rates by MDSA that were needed to explain the Barbados anomaly.

M.M.J.W.v.H., Q.L., A.S.-L., T.R., and M.S.J. designed research; M.M.J.W.v.H., Q.L., and M.S.J. performed research; Q.L., A.S.-L., T.R., C.A.C., R.P.F., J.E.M., and J.-B.W.S. contributed new reagents/analytic tools; M.M.J.W.v.H., Q.L., J.B.L., T.R., N.M.M., D.M., and M.S.J. analyzed data; and M.M.J.W.v.H., Q.L., A.S.-L., J.B.L., T.R., C.A.C., R.P.F., J.E.M., N.M.M., P.H., D.M., J.-B.W.S., and M.S.J. wrote the paper.

The University of Copenhagen (UCPH) has filed a patent application on quantifying atmospheric methane removal by chlorine atoms from iron photochemistry (inventors M.M.J.W.v.H. and M.S.J.). All other authors declare they have no competing interests.