Meteor-Ablated Aluminum in the Mesosphere-Lower Thermosphere

The first global atmospheric model (WACCM-Al) of meteor-ablated aluminum was constructed from three components: The Whole Atmospheric Community Climate Model (WACCM6); a meteoric input function for Al derived by coupling an astronomical model of dust sources in the solar system with a chemical meteoric ablation model; and a comprehensive set of neutral, ion-molecule and photochemical reactions relevant to the chemistry of Al in the upper atmosphere. The reaction kinetics of two important reactions that control the rate at which Al + ions are neutralized were first studied using a fast flow tube with pulsed laser ablation of an Al target, yielding k (AlO + + CO) = (3.7 ± 1.1) × 10 −10 and k (AlO + + O) = (1.7 ± 0.7) × 10 −10 cm 3 molecule −1 s −1 at 294 K. The first attempt to observe AlO by lidar was made by probing the bandhead of the B 2 Σ +

with a time constant of ∼300 μs at 85 km. Thus, unlike the major meteoric metals Fe, Mg, and Na which occur as layers of neutral metal atoms between about 80 and 105 km, aluminum is likely to occur predominantly as a layer of AlO. There are two reasons for this conjecture. First, atomic Si is the only other major meteoric species which undergoes a fast bimolecular reaction with O 2 , and a detailed model of silicon chemistry predicts that this element occurs as a layer of SiO rather than Si in the MLT . Second, solar-pumped fluorescence from the AlO(B 2 Σ + − X 2 Σ + ) band has been observed when tri-methyl aluminum (TMA) is released in the MLT during twilight (Golomb et al., 1967;Johnson, 1965;Rosenberg et al., 1964). Emission from the same AlO band was also observed during entry of the very bright Benešov bolide over the Czech Republic (Borovička & Berezhnoy, 2016).
The only aluminum species which has so far actually been observed in the background atmosphere is the 27 Al + ion, measured using rocket-borne mass spectrometry (Grebowsky & Aikin, 2002;Kopp, 1997;Krankowsky et al., 1972). The Al + /Fe + ratio between 90 and 100 km was found from a series of rocket flights to be 0.022 ± 0.005 , which is reasonably close to the estimated Al/Fe meteoric ablation ratio of 0.037 (Carrillo-Sánchez et al., 2020). We have recently carried out a study of the kinetics of the pertinent neutral (Gómez Martín, Daly, et al., 2017;Mangan et al., 2020) and ion-molecule  reactions that aluminum is likely to undergo in the MLT. These studies, along with electronic structure theory calculations to elucidate likely reaction pathways, has enabled the chemical network shown in Figure 1 to be constructed. The reactions that we have measured previously are indicated with blue arrows.
In terms of ion-molecule chemistry, Al + mostly reacts with O 3 in the MLT ( Figure 10 in Daly et al. [2019]) to produce AlO + . In Section 2.1 of the present study, we describe an experimental study to measure the rate coefficients for the reactions of the AlO + ion with O and CO (red arrow in Figure 1): (note that the reaction numbering follows the complete list of reactions in Table 1). These two highly exothermic reactions (the reaction enthalpies are calculated using the electronic structure method discussed in Section 2.2) control the balance between ionized and neutral aluminum because they reduce AlO + to Al + , which can only undergo slow dielectronic recombination with electrons (see Figure 1).
In terms of neutral chemistry, the measured reaction kinetics indicate that AlO will initially form OAlO 2 and AlCO 3 (see Figure 11 in Mangan et al., 2020). However, AlCO 3 may then react exothermically with O 2 to form OAlO 2 , which in turn is likely to react with H to produce AlOH, as shown in Figure 1. Unlike other metal hydroxides such as FeOH (Self & Plane, 2003), NaOH (Gómez Martín, Daly, 2017), and CaOH , AlOH is stable with respect to reaction with H and O atoms  and is therefore likely to be a major Al reservoir. In Section 2.2, we use electronic structure theory calculations to explore these pathways for converting AlO to AlOH.
In fact, it appears that the only process which can recycle AlOH to AlO directly is photolysis. The excited electronic states of AlOH have been studied in some detail by Trabelsi and Francisco (2018) (in order to explain the observed ratio of AlO to AlOH in the interstellar medium). Using high level coupled cluster theory calculations, they showed that the two photolysis channels should have almost identical thresholds around 225 nm . Note that any Al produced via channel R17a will immediately be oxidized to AlO via reaction for photolysis reactions; R1. In Section 2.3, the photodissociation rate of AlOH in the MLT is estimated.
In Section 3, we describe a set of lidar observations of the expected AlO layer. The peak absorption cross section of AlO in the B-X band at 484.23 nm was measured in our laboratory to be σ(298 K) = (6.7 ± 1.6) × 10 −15 cm 2 molecule −1 (Gómez Martín, Daly, et al., 2017). This cross section is unusually large for a molecular diatomic transition, and is only a factor of 80 smaller than the cross section for atomic Fe at 372 nm used for lidar measurements of the Fe layer in the MLT. It is worth emphasizing that although chemiluminescence from FeO and NiO has been observed in the nightglow spectrum (Evans et al., 2011;Saran et al., 2011), no molecular metallic species has been actively detected by resonance lidar. The lidar results are then compared with an estimate of the AlO peak density determined from the lifetime of the AlO trails produced by TMA releases.
In Section 4, we incorporate into a whole atmosphere chemistry-climate model the aluminum chemistry network shown schematically in Figure 1, together with a meteoric input function for Al (Carrillo-Sánchez et al., 2020). The model simulations are then compared with observations of Al + and AlO.

Experimental Study of AlO + Reaction Kinetics
Reactions R22-R23 were studied in a stainless-steel fast flow tube which has been described in detail previously Daly et al., 2019). At the upstream end of the tube, a pulsed Nd:YAG laser (Continuum Surelite) was used to ablate Al + ions from a rotating Al rod, which were then entrained in a carrier gas flow of He (mass flow rate ranging from 3.3 to 3.5 standard liters min −1 ). O 3 was added at a fixed injection point 19 cm downstream of the Al rod to produce AlO + via reaction R21 . Atomic O or CO was then added further downstream via a sliding injector. At the downstream end of the flow tube, after a reaction time of several milliseconds, Al + ions were detected with a quadrupole mass spectrometer (Hiden Analytical, model HPR60) operating in positive ion mode. A roots blower backed by a rotary pump provided a range of flow velocities from 48 to 76 m s −1 , at the constant pressure of 1.0 Torr which was used in these experiments. The resulting reaction times after injection of O or CO ranged from 7.5 to 8.0 ms. All experiments were conducted at 294 K.
O 3 was generated by passing O 2 through a high voltage corona discharge in a commercial ozonizer, with its concentration measured spectrophotometrically at 253.7 nm (provided by a Hg pen lamp) in a 19 cm pathlength optical cell. The O 3 absorption cross section used was 1.16 × 10 −17 cm 2 molecule −1 (Molina & Molina, 1986). Atomic O was generated through microwave discharge of N 2 (McCarroll cavity, Opthos Instruments Inc.), followed by titration with NO before injection into the flow tube through the sliding injector (Self & Plane, 2003 Units: s −1 for photolysis reactions; cm 3 molecule −1 s −1 for bimolecular reactions; cm 6 molecule −2 s −1 for termolecular reactions. b Gómez Martín, Daly, et al., 2017. c Mangan et al., 2020. d set to a collision frequency of 2 × 10 −10 (T/300) 1/6 cm 3 molecule −1 s −1 , scaled by a statistical electronic branching factor (see text). e Calculation from electronic structure theory (Section 2.3); note that any Al produced by R17a will immediately be oxidized to AlO by reaction R1. f Set to the Langevin collision frequency, scaled by a statistical electronic branching factor (see text). g Daly et al., 2019. h (Self & Plane, 2003).
Materials: carrier gas He (99.995%, BOC gases) was purified through a molecular sieve at 77 K before flow tube entry; N 2 (99.9999%, Air products), O 2 (99.999%, Air products) and CO (99.5% pure, Argo International) were used without further purification; NO (99.95%, Air products) was purified via three freeze-pump-thaw cycles before dilution in He.  Figure 2 illustrates how this fraction increases as a function of [CO], due to R23 converting AlO + back to Al + .

Reaction of AlO + + CO
The flow tube kinetics are complicated by the additional reactions of AlO + with O 3 and O 2 , as well as diffusional loss of the ions to the flow-tube walls. A kinetic model of the flow tube was therefore used to determine the rate coefficient k 23 . The model uses a set of ordinary differential equations (ODEs) to describe the time-dependent variation of Al + , AlO + , and AlO 2 + down the length of the flow tube. The model is described in detail elsewhere . The first-order wall loss rate (k diff ) for Al + was measured to be 655 ± 15 s −1 at 294 K and 1 Torr . k diff for AlO + and AlO 2 + were calculated to be 650 and 649 s −1 , respectively, from the long-range ion-induced dipole forces between these ions and the He bath gas . The rate coefficients and branching ratios for the reactions of Al + and AlO + with O 2 and O 3 have been measured previously by Daly et al. (2019), and are listed in Table 1.
A value for k 23 was obtained by independently fitting the model to each experimental data point in Figure 2, and then calculating an overall mean value and standard deviation of k 23 = (3.7 ± 1.1) × 10 −10 cm 3 molecule −1 s −1 at 294 K. The model run using this result is shown as the solid line in Figure 2 (the dashed lines indicate the uncertainty in k 23 ), and clearly provides a satisfactory fit to the experimental data.  , the reaction of the Al + .N 2 cluster ion with O did not have to be accounted for in the model (the source of N 2 is the microwave discharge) because the reaction between Al + and N 2 is very slow . Figure   absence of O, yielding k 22 (294 K) = (1.7 ± 0.7) × 10 −10 cm 3 molecule −1 s −1 .

Reaction of AlO + + O
The dashed lines illustrate the model fit with k 22 set to its upper and lower limits at the 1σ uncertainty level.

Neutral Al Chemistry
In order to explore the likely balance between AlO and AlOH in the MLT, we examine here the pathways from OAlO 2 and AlCO 3 to AlOH (Figure 1). H and H 2 O have similar concentrations between 80 and 90 km , and so direct conversion of OAlO 2 to AlOH (R10), and indirect conversion via Al(OH) 2 (R11 + R12), need to be considered: Although these reactions are highly exothermic, it is important to determine whether there are any substantial energy barriers on the potential energy surfaces (PESs) which link the reactants to the products. Electronic structure calculations were used to do this. The geometries of the Al-containing molecules were first optimized at the B3LYP/6-311+g(2d, p) level of theory within the Gaussian 16 suite of programs (Frisch et al., 2016), and then more accurate energies determined using the Complete Basis Set (CBS-QB3) method (Montgomery et al., 2000). The PESs for R10, R11, and R12 are illustrated in Figure 4, which also shows the geometries of the stationary points on each surface. The Cartesian coordinates, rotational constants, vibrational frequencies and heats of formation of the relevant molecules are listed in Table S1.
All three reactions exhibit deep wells on their PESs, corresponding to very stable intermediates. However, there are no barriers above the reactant entrance channel energies. Hence, at the low pressures of the MLT these intermediates will not be stabilized by collisional quenching with air molecules, and the bimolecular products should form with rate coefficients that are close to their collision frequencies and have small temperature dependences. Interestingly, the reaction between OAlO 2 and H can take place on surfaces of either singlet or triplet spin multiplicity. Although the singlet surface has a deeper well corresponding to singlet HOAlO 2 , spin conservation means that this species will dissociate to AlOH( 1 A′) and electronically excited O 2 ( 1 Δ g ).
In the case of AlCO 3 , the most likely reaction is with O 2 to form OAlO 2 , although reaction with H to make AlOH directly, or indirectly with H 2 O via Al(OH) 2 , are also exothermic:  (Table S1).
The PESs for these three reactions ( Figures S1-S3) show that there are no barriers, so these reactions should also be close to their collision frequencies. The same considerations apply to exothermic reactions R8, R9, and R13. In order to assign rate coefficients to R8-R16, we assume a typical collision frequency of 2 × 10 −10 cm 3 molecule −1 s −1 (T/300) 1/6 cm 3 molecule −1 s −1 , and multiply this by a statistical factor if the combination of reactant spins leads to a multiplicity of PESs which exceeds that of the products (Smith, 1980). For example, for R15 the products are both singlets, and the reactants are both doublets, so the statistical factor is (1 × 1)/(2 × 2) = 0.25. These rate coefficients are listed in Table 1.

Photochemistry of AlOH
We have shown previously that the observed growth of Fe on the underside of the mesospheric Fe layer at sunrise is most probably due to the photolysis of the reservoir species FeOH, which has a relatively large photolysis rate in the MLT of J(FeOH) = (6 ± 3) × 10 −3 s −1 (Viehl et al., 2016). Here, we use the quantum chemistry method that we used previously for FeOH (Viehl et al., 2016) and NiOH  to estimate J(AlOH). First, the geometry of AlOH was optimized at the B3LYP/6-311+g(2d, p) level of theory (Frisch et al., 2016). Second, the vertical excitation energies and transition dipole moments for transitions from the AlOH ground state to the first 50 electronically excited states were calculated using the time-dependent density function theory (TD-DFT) method (Bauernschmitt & Ahlrichs, 1996).
The resulting absorption spectrum is plotted in Figure 5, which shows that the threshold for photodissociation occurs close to the peak of a strong near-UV absorption band peaking at 229 nm. If absorption at wavelengths shorter than 225 nm causes photodissociation to either Al + OH or AlO + H (Trabelsi & Francisco, 2018), then convolving the AlOH cross section up to this threshold with the solar actinic flux from the semi-empirical SOLAR2000 model (Tobiska et al., 2000) (averaged over a solar cycle), yields J(AlOH) = 3.3 × 10 −3 s −1 in the MLT.

Al Ion-Molecule Chemistry
The ionization energies of AlO and AlOH are 9.82 eV (Clemmer et al., 1992) and 8.89 eV (Sikorska & Skurski, 2009), respectively. These are both lower than the ionization energy of O 2 (12.07 eV), which means that both AlO and AlOH should charge transfer with ambient E region O 2 + ions (R18 and R19). However, the lower ionization energy of NO (9.26 eV) means that only AlOH will charge transfer with ambient NO + (R20). The rate coefficients for these reactions are set to their Langevin capture rates, increased to account for the significant dipole moments of AlO (4.45 D [Bai & Steimle, 2020]) and AlOH (0.97 D [Sikorska & Skurski, 2009]) using the statistical adiabatic model of Troe (1985). These capture rates are then multiplied by a statistical factor to take account of the spin multiplicities of reactants and products.
Al + reacts most rapidly with O 3 (R21 in Table 1) throughout the MLT . AlO + is then most likely to react with O and be reduced back to Al + (R22, see Section 2.1). However, AlO + can also recombine with N 2 (R33 in Table 1). The rate coefficient k 33 was calculated using the version of Rice Ramsperger Markus Kassel (RRKM) theory described in Daly et al. (2019). The relevant molecular parameters are listed in Table S4. This reaction is reasonably fast because the AlO + .N 2 cluster ion is bound by 106 kJ mol −1 . It is then likely to react with O to form the weakly bound Al + .N 2 ion, which can ligand switch with CO 2 and H 2 O to form more stable Al + .CO 2 and Al + .H 2 O cluster ions . Note that all three of these cluster ions can also form directly through the recombination of Al + with N 2 , CO 2 or H 2 O (R27-R29), though only the Al + + N 2 reaction is within two orders of magnitude of reaction with O 3 (R21) . The three cluster ions can then be converted to AlO + by reaction with O (R30-R32).
The rate coefficients of all the relevant bimolecular ion-molecule reactions which have not been measured (black arrows in Figure 1) are set to their Langevin capture rates (Smith, 1980  ions can all undergo dissociative recombination with electrons (R35). These reactions are all set to the rate coefficient measured for FeO + + e − (Bones et al., 2016), based on the observation that dissociative recombination reactions of small molecular ions nearly all have rate coefficients within a factor of 2 of 3 × 10 −7 cm 3 molecule −1 s −1 (Florescu-Mitchell & Mitchell, 2006).

Permanent Removal of Al Species
Reaction R36 in Table 1 is a set of polymerization reactions which account for the permanent loss of the significant neutral Al-containing molecules AlO, AlOH and, to a lesser extent, Al(OH) 2 (see Section 4) to form meteoric smoke particles (MSPs). We have used this type of reaction in previous models of the Na , K (Plane et al., 2014), Fe , Mg (Langowski et al., 2015), SiO , Ca (Plane et al., 2018), and Ni  layers. In this case, k 36 is set to 5.8 × 10 −8 cm 3 s −1 , which is ∼80 times larger than a typical dipole-dipole capture rate for these metallic molecules. This factor allows for the Al-containing reservoir species to polymerize with other metal-containing molecules produced by meteoric ablation (e.g., FeOH and Mg(OH) 2 ), whose concentration will be around 80 times higher because the elemental ablation ratio of Al atoms to the sum of Na + Fe + Mg + Si + Ni + Al atoms is 1/81.2 (Carrillo-Sánchez et al., 2020).

Lidar and Calibration Cell Setup
The absolute absorption cross section of AlO at the bandhead of the B 2 Σ + (v′ = 0) − X 2 Σ + (v′′ = 0) transition at λ air = 484.23 nm is σ(298 K, 1 hPa) = (6.7 ± 1.6) × 10 −15 cm 2 molecule −1 (0.003 nm resolution) (Gómez Martín, Daly, et al., 2017). Because this cross section is unusually large for a diatomic molecule, we carried out a lidar campaign to determine if an AlO layer could be detected. Soundings were performed at Kühlungsborn, Germany (54°N, 12°E) for three nights during January 2016 and three nights in April 2017, yielding ∼20 h of integration time. Details of the lidar system are given by Gerding et al. (2019); the instrument is a modification from an earlier twin dye laser design (Alpers et al., 1996;Gerding et al., 2000). Laser emission at 484.23 nm was produced using a XeCl excimer laser at 308 nm (repetition rate = 30 Hz) to pump a dye laser with Coumarin 102 dye dissolved in methanol, producing laser radiation over the 455-495 nm spectral range.
A small-scale version of the flow tube used by Gómez Martín, Daly et al. (2017) was installed next to the lidar as an AlO calibration cell, both to check the laser wavelength before atmospheric measurements and then to avoid drift away from the AlO bandhead during operation. AlO was produced in the cell by laser ablation of a rotating Al rod, using 532 nm light that was beam-split from a Nd:YAG laser in the co-located Rayleigh-Mie-Raman (RMR) lidar (Gerding et al., 2016). The Al was entrained in a flow of N 2 (total pressure = 2.1 Torr), and a trace of O 2 added N 2 downstream to make AlO via reaction 1. A quartz fiber was used to guide the 484 nm laser light from the AlO lidar to the calibration cell, and laser induced fluorescence detected with a photomultiplier orthogonal to the laser beam. The dye laser was scanned in 1 pm intervals to find the peak of the AlO bandhead. A flip mirror was used to alternately direct the dye laser to the calibration cell or to the optics in the lidar transmitter. Figure 6a shows the integrated lidar backscatter profile at 484.23 nm (blue line), summed over the three sounding nights during April 2017. The background noise level, which was determined by averaging the signal from 120 to 150 km (dashed line in Figure 6a), has been subtracted. The RMR lidar (green line in Figure 6), which operated simultaneously alongside the AlO resonance lidar, was used to provide an off-resonance measurement (at 532 nm) since no off-resonance measurements were taken with the resonance lidar (which was set to the AlO bandhead). Both profiles showed a monotonic decay of the Rayleigh scatter into the background noise. The 484 nm Rayleigh scatter was detected well above 80 km where an AlO layer would be expected (Figure 6b), based on the metal atom layers (Gerding et al., 2019;Plane, 2003).

Observations at 484 nm
The Rayleigh backscatter was then extrapolated from 80 km to higher altitudes (purple lines in Figures 6a  and 6b) and subtracted from the backscatter signal to yield the residual signal (black line in Figure 6c). No obvious resonance layer was detected over the observation period; application of Poisson statistics shows that an AlO resonance signal was not present above the the 3σ photon noise threshold (red line in Figure 6c) (Gerrard et al., 2001). Nevertheless, an upper limit for the AlO density can be estimated. A Gaussian profile for the AlO layer was assumed, extending from 85 to 100 km with a peak at 90 km (analogous to other metal layers [Plane, 2003]), and fitted to the residual signal. Adapting the work of Tilgner and von Zahn (1988), the upper limit to the AlO density, n z (AlO), is then given by: where n zr (air) is the air density at the reference altitude (3.6 × 10 17 cm −3 ) from NRLMSISE-00 (Picone et al., 2002); σ Ray (7.6 × 10 −27 cm 2 molecule −1 ) and σ res ((6.7 ± 1.6) × 10 −15 cm 2 molecule −1 [Gómez Martín, Daly, et al., 2017]) are the Rayleigh and effective resonance AlO cross sections (the backscatter cross sections are these values divided by 4π); z(AlO) is the assumed altitude for the AlO peak (90 km), z r the reference altitude of 30 km; C(AlO) and C(air) are the AlO resonance and Rayleigh photon counts after the background noise is subtracted (20 and 1.3 × 10 6 counts, respectively); and Tr (z r , z) is the transmission (assumed to be PLANE ET AL. 10.1029/2020JA028792 9 of 15 1) of the atmosphere between z r and z at the laser pulse wavelength. This yields an AlO detection limit of 60 cm −3 .

Al Releases in the MLT
TMA grenade releases from rocket payloads in the MLT generate visible chemiluminescence (Golomb et al., 1967;Roberts & Larsen, 2014), which was proposed to arise from the radiative recombination reaction (Golomb & Brown, 1976;Rosenberg et al., 1964): R37 is sufficiently exothermic to produce emission at wavelengths longer than 306 nm, as we have shown recently in the laboratory . Note that the reaction between Al and H 2 O proposed by Gole and Kolb (1981) is not required to explain the chemiluminescence; in any case, any Al will preferentially react with O 2 (R1). The OAlO product is then recycled to AlO by reaction with O: which proceeds close to the capture rate , so that AlO is in a large excess over OAlO and the intensity of the chemiluminescence is a marker for the AlO concentration. Roberts and Larsen (2014) reported that the chemilumiscence intensity decayed with an e-folding lifetime of around 29 min between 90 and 100 km i.e. the first-order removal for AlO into a long-lived reservoir is ∼6 × 10 −4 s −1 . The rate of injection of Al atoms into the MLT has recently been estimated to be 3 × 10 −3 cm −3 s −1 (Carrillo-Sánchez et al., 2020); since the Al will immediately be oxidized by O 2 to AlO, this represents the injection rate of fresh AlO. Balancing injection against removal, the steady-state concentration of AlO should then be ∼5 cm −3 . This is 1 order of magnitude lower than the upper limit for AlO determined from the lidar observations in Section 3.1. Note that this estimate of the AlO density is during nighttime, when these rocket release experiments were conducted.

WACCM-Al Set up
The Al reactions in Table 1 were imported into the Whole Atmosphere Community Climate Model (WAC-CM6), which uses the framework developed from the second iteration of the fully coupled Community Earth System Model (CESM2) (Gettelman et al., 2019). WACCM6 has a vertical extension from the Earth's surface to the lower thermosphere at ∼140 km. Although the model can be nudged by a reanalysis data set, as we have done with other meteoric metals where measurements are available for comparison (Plane et al., , 2018, for the present study we used a free-running version of WACCM6 with a reduced tropospheric chemical mechanism. The model has a horizontal resolution of 1.9° latitude × 2.5° longitude, and 70 vertical model layers (∼3 km vertical resolution in the MLT region). This version of WACCM6 with Al chemistry is termed WACCM-Al. The full set of Fe reactions in WACCM-Fe Viehl et al., 2016) was also included in order to compare model simulations with measurements of Al + and Fe + in the MLT. The model simulations were performed from 1979 to 2014, using the standard WACCM6 initialization conditions file (Danabasoglu et al., 2020). Here we focus on a decade of model output from 2004 to 2014, which is sufficiently long to produce a climatology of the Al species.

Al Meteoric Input Function
The global average injection profiles of Al and Fe are illustrated in Figure 7. ABLation MODel (CABMOD-3), which simulates the ablation of the major meteoric elements from an individual dust particle (Carrillo-Sánchez et al., 2020), with the Zodiacal Cloud Model (ZoDY) which provides the mass, velocity and radiant distributions of particles entering Earth's atmosphere from Jupiter Family Comets, the asteroid belt, and long-period Halley-Type comets (Nesvorný et al., 2011). The contributions from these different sources are weighted using the procedure in Carrillo-Sánchez et al. (2016). The upper peak in the Al injection profile arises from fast meteors (mostly from Halley-Type comets), and the lower peak from slow meteors (mostly from Jupiter Family comets).
Note that both injection profiles in Figure 7 have been reduced by a factor of 5 from the profiles in Carrillo-Sánchez et al. (2020). This accounts for the fact that global models such as WACCM underestimate the vertical transport of minor species in the MLT, because short wavelength gravity waves are not resolved on the current horizontal grid scale of the model (∼220 km). These subgrid waves contribute to vertical chemical and dynamical transport of constituents while dissipating, and this can exceed transport driven along mixing ratio gradients by the turbulent eddy diffusion produced once the waves break (Gardner et al., 2017). Because these additional vertical transport mechanisms are underestimated, the MIF of each metal needs to be reduced in order to correctly simulate the observed absolute metal density (Plane et al., 2018). Note that this 5-fold reduction in the Al MIF was not applied in Section 3.2 when estimating the steady-state AlO concentration using the observed AlO decay rate after a rocket release. This is because the rocket experiments are in the real atmosphere. The Al MIF in WACCM is then set to vary with season and latitude in the same way as the Fe MIF  , that is, an autumnal maximum and vernal minimum, increasing from essentially no variation at the equator to ±30% at the pole, with the same annual average input at all latitudes. PLANE ET AL.   Figure 8 shows the annual average vertical profiles of the major Al species at 54 o N, the latitude of the lidar observations. As expected, Al + is the dominant species above 95 km. AlO and AlOH then occur in layers that peak around 89 and 86 km, respectively. Al(OH) 2 is also significant below 80 km once the atomic H concentration decreases significantly , so that reaction R12 becomes very slow. Below 92 km, most of the Al is tied up as Al-containing polymers, which represent a surrogate for MSPs (see Section 2.5). The uncertainty in the modeled balance between AlO and Al + arising from the experimental uncertainties in k 22 and k 23 (Section 2.1) was estimated using a Monte Carlo analysis. The uncertainty in [AlO] increases from 12% to 18% between 87 and 110 km, but that in [Al + ] is ≤ 5% because this is the dominant gas-phase species above 87 km (Figure 8). Figure 9 (left panel) compares the vertical profile of Al + simulated by WACCM-Al with the geometric mean profiles from a set of eight mid-to high-latitude rocket-borne mass spectrometric measurements by Kopp and co-workers (Kopp, 1997;Kopp et al., 1984Kopp et al., , 1985aKopp et al., , 1985bMeister et al., 1978). Details of these flights are provided in Table S5. The model results are the annual average simulated Al + profile at 0 LT for 54 o N. The observed Al + layer peaks around 92-94 km, with a geometric mean density of 40 cm −3 and geometric standard deviation from 20 to 100 cm −3 . The modeled layer peaks at 93 km, with a density close to 100 cm −3 . Given the paucity of observations, this level of agreement is satisfactory. Because the reaction of AlO + with O (R22) is relatively fast (Section 2.1.2) and O is a major species above 84 km, Al + is the major Al-containing ion by 2-5 orders of magnitude between 85 and 110 km. Figure 9 (right panel) shows that the rocket-measured Fe + :Al + ratio is also satisfactorily modeled between 86 and 104 km. The ratio is very close to the CABMOD-Zo-Dy estimate of the relative meteoric inputs, which is a factor of 2.8 larger than the CI ratio of the two metals. Figure 10 shows the diurnal variation of Al + , AlO and AlOH as a function of height during April at 54 o N, in order to compare with the lidar measurements described in Section 3.1. The diurnal variation of the vertical column densities of these species is shown in Figure S4. As expected, Al + peaks between 13 and 17 UT because of the daytime increase in the concentrations of the lower E region ions NO + and O 2 + , which charge transfer with AlO and AlOH (R18-R20).

Model Results
More interesting is the diurnal behavior of the neutral species AlO and AlOH, which are essentially anticorrelated: AlO peaks during daytime, and AlOH at night. This behavior is caused by the photolysis of AlOH (R17) to produce AlO either directly or via Al. The result is that AlO varies between 10 and 20 cm −3 at night, but it increases to over 60 cm −3 between 13 and 20 UT. The nighttime level is consistent with the upper limit of 60 cm −3 determined from the lidar observations (Section 3.1), and also with the concentration of ∼5 cm −3 that is inferred from the Al rocket release experiments (Section 3.2). Figure 11 illustrates the variation with latitude and month of the vertical column densities of Al + , AlO, and AlOH. Al + shows little seasonal variation at low latitudes, but a three-fold increase between winter and summer at mid-to high-latitudes, reflecting the change in ambient lower E region ionization. AlOH also demonstrates a strong (though opposite) annual cycle at high latitudes, increasing by a factor of ∼6 from a mid-summer minimum in the continuously sunlit polar region to a mid-winter maximum in polar night. In contrast, AlO exhibits a semi-annual cycle at mid-to high-latitudes, peaking at the equinoxes. The reason is that after polar night, during which AlO is very low because most of the neutral Al is in the form of AlOH, photolysis causes a spring-time increase in AlO by a factor of ∼3. However, moving into summer the AlO is reduced again by increased charge transfer with O 2 + (R18), causing Al + to peak. The situation then reverses in the autumn. Note that the AlO is up to a factor of 1.3 times higher at the autumnal compared with the vernal equinox, because of the autumnal peak in the MIF . Figure S5 illustrates the seasonal/latitudinal variation of the centroid height and root-mean-square (RMS) width of the AlO layer. Although the layer mostly peaks around 90 km, at high latitudes during polar night the peak increases to 98 km because AlOH is essentially a sink for neutral Al species below this in the absence of sunlight. In contrast, the mid-summer AlO layer at high latitudes still peaks around 90 km because now the removal of AlO is via charge transfer at higher altitudes. The RMS layer width is on average around 5 km, with a mid-summer minimum at polar latitudes of 3.4 km because of the ionization of the top-side of the AlO layer.

Conclusions
In this study, we describe a comprehensive Al chemistry network, constructed from a set of neutral and ion-molecule reactions measured previously in our laboratory Mangan et al., 2020), as well as the reactions of AlO + with O and CO (R22 and R23) reported as part of the present study. Additional reaction rate coefficients are estimated by using electronic structure theory to explore the relevant PESs. The Al reaction network was then incorporated into the WACCM chemistry-climate model, along with a new MIF for Al (Carrillo-Sánchez et al., 2020).
We also report the first attempt, to our knowledge, to directly observe the AlO layer in the MLT. Although the lidar observations did not detect a layer, an upper limit of only 60 cm −3 for the AlO density was determined. This sets an important benchmark for future observations. A rough estimate for AlO of around 5 cm −3 was obtained from the rate of decay of AlO chemiluminescence from rocket-borne grenade releases. Both of these types of atmospheric measurements apply to nighttime. However, the WACCM-Al model indicates that AlO should be a factor of ∼6 times higher during daytime, because of photolysis of AlOH, which is the other major neutral Al-containing molecule. Of course, this result depends on the accuracy of the calculated photolysis rate of AlOH (Section 2.3), and it is essential that this is measured in the future. Lidar measurements during twilight, when photolysis of AlOH in the MLT is still occurring but the solar terminator is above the troposphere so that the amount of scattered sunlight is reduced, would offer the best chance of detecting AlO. This is particularly the case at high latitudes during spring or autumn (e.g., at 69°N, 3 h of twilight measurements could be made on Julian days 112 and 253), when the AlO density should also be high ( Figure 11).

Data Availability Statement
The version cesm2_1_3 model and input data are provided by the National Center for Atmospheric Research (http://www.cesm.ucar.edu/models/cesm2/release_download.html). The WACCM-Al and WACCM-Fe models and output are archived in the Petabyte Environmental Tape Archive and Library at the University of Leeds via https://petal.leeds.ac.uk/. The data are available at http://doi.org/10.5281/zenodo.4066748.