Benjamin
Yang
*ab,
Forwood C.
Wiser
c,
V. Faye
McNeill
bc,
Arlene M.
Fiore
d,
Madankui
Tao
abd,
Daven K.
Henze
e,
Siddhartha
Sen
f and
Daniel M.
Westervelt
*a
aLamont-Doherty Earth Observatory of Columbia University, Palisades, NY, USA. E-mail: benjamin.yang@columbia.edu; danielmw@ldeo.columbia.edu
bDepartment of Earth and Environmental Sciences, Columbia University, New York, NY, USA
cDepartment of Chemical Engineering, Columbia University, New York, NY, USA
dDepartment of Earth, Atmospheric, and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA, USA
eDepartment of Mechanical Engineering, University of Colorado Boulder, Boulder, CO, USA
fMicrosoft Research, New York, NY, USA
First published on 2nd November 2023
Detailed chemical mechanisms are computationally challenging to include in large-scale chemical transport models such as GEOS-Chem. Employing a graph theory-based automated model reduction (AMORE) algorithm, we developed a new reduced (12 species and 23 reactions) gas-phase isoprene oxidation mechanism. We performed GEOS-Chem simulations for a full year (June 2018–May 2019) with the default (BASE) and AMORE version 1.1 isoprene mechanisms at 2° × 2.5° horizontal resolution globally and 0.25° × 0.3125° resolution over the eastern United States (EUS). Additionally, we conducted BASE and AMORE sensitivity simulations in which biogenic isoprene and anthropogenic emissions were sequentially set to zero in the model. For the entire year simulated, GEOS-Chem was faster by 10% in total and 25% in the chemical reaction solver (KPP) with the AMORE mechanism. Evaluating GEOS-Chem against surface observations from the Air Quality System (AQS) and Interagency Monitoring of Protected Visual Environments (IMPROVE) networks as well as satellite columns from the Tropospheric Monitoring Instrument (TROPOMI) and Cross-track Infrared Sounder (CrIS), our results show comparable accuracy in BASE and AMORE nested-grid simulations of air pollutants, with annual mean model bias changes of 1% for both surface PM2.5 and O3 over the EUS. From the sensitivity simulations, we find that US biogenic isoprene contributes to 8–9% of surface PM2.5 and 3–4% of surface O3 on average in summer over the EUS. This study indicates that AMORE is an attractive option for future GEOS-Chem modeling studies, especially where detailed isoprene chemistry is not the focus.
Environmental significanceAtmospheric oxidation of isoprene leads to the secondary formation of air pollutants, namely fine particulate matter and ozone, which affect health and climate. The full isoprene oxidation mechanism is highly complex, involving hundreds of species and over a thousand reactions. To accurately and efficiently model atmospheric chemistry, the optimal balance between small mechanism size and accuracy should be achieved. We find that using our automated model reduction (AMORE) isoprene mechanism over the default mechanism in the GEOS-Chem chemical transport model not only speeds up simulations but also roughly maintains the accuracy of simulated air pollutant concentrations compared with surface and satellite-based observations. Our AMORE approach has the potential to be applied to other chemical mechanisms and models. |
Isoprene (C5H8) is the predominant non-methane biogenic volatile organic compound (BVOC), representing about 50% of 1 Pg of total global annual BVOC emissions.5 Most isoprene-emitting plant species are located in the humid tropics, and, in temperate regions, oak and aspen trees emit significant amounts of isoprene.6 Monoterpenes and sesquiterpenes are other important BVOCs, constituting 15% and 3% of total global BVOC emissions.5 The southeastern United States (US) is an extensively-studied isoprene hotspot with significant anthropogenic emissions. There, isoprene is primarily emitted during the warm, growing season and dominates hydroxyl radical (OH) reactivity in summer, while monoterpenes remain more constant and dominate O3 and nitrate radical (NO3) reactivity throughout the year.7 In GEOS-Chem, isoprene is mostly oxidized by OH (85%), followed by O3 (11%) and NO3 (4%) pathways, over the southeastern US.8
Atmospheric photooxidation of VOCs including isoprene by OH produces an organic peroxy radical that converts nitric oxide (NO) to nitrogen dioxide (NO2).9 An organic oxy radical reacts with oxygen to form hydroperoxyl radical (HO2), and the HO2 additionally converts NO to NO2.9 In the presence of sunlight, photolysis of NO2 forms O3, which means that both isoprene and total NOx (NO + NO2) levels partially control O3 production. Formaldehyde (HCHO) and other oxygenated VOCs are formed in the oxidation process. Furthermore, biogenic isoprene and monoterpenes are responsible for 60% of organic aerosol (OA), a major component of PM2.5, over the southeastern US.10 Isoprene and its intermediates participate in an elaborate series of chemical reactions to form lower volatility products that partition into OA depending on the ambient conditions.11 For example, isoprene species react with OH and HO2 to form isoprene epoxydiols (IEPOX) under low-NOx conditions.12 It is estimated that IEPOX, organonitrates, and tetrafunctional compounds each contribute to one-third of global isoprene secondary organic aerosol (SOA).8 Yet, past studies have indicated many unknowns in isoprene chemistry, particularly the fate of isoprene nitrates.13–15
In atmospheric chemistry modeling, we need innovative ways for representing complex, nonlinear, multiphase, and multiscale chemical processes. Currently, the master chemical mechanism (MCM version 3.3.1) for isoprene oxidation involves 602 species and 1926 reactions, derived from theoretical studies and chamber experiments.16 Adding 86 species and 358 reactions to a detailed, comprehensive mechanism of isoprene and terpene chemistry (MOZART) was shown to decrease the daily maximum surface O3 bias of the CESM/CAM-chem model generally by 3–4 ppb in the eastern US, relative to the previous, smaller MOZART mechanism, and increased computation time by 50%.15 However, it is not always practical to capture the detailed chemistry in large-scale atmospheric chemical transport models (CTMs) including GEOS-Chem or chemistry-climate models (CCMs) due to excessive computational costs. Consequently, CTMs and CCMs often use reduced mechanisms while striving for accuracy within acceptable ranges as a trade-off. Even in reduced form, chemistry involving isoprene is one of the most computationally-intensive components in these models.
To date, chemical mechanism reduction has been mostly performed manually by expert air quality scientists and software developers through techniques such as chemical lumping and empirical parameterization. Condensing the mechanism may occur in several stages including isomer grouping, applying steady-state approximation, and lumping minor and latter pathways together.17 Species are also lumped, resulting in different nomenclature across mechanisms employed in CTMs.8 The reduced mechanisms typically lack flexibility to adapt to new environmental conditions and require significant labor, time, and resources to make further updates. Updates are infrequent and susceptible to errors. Therefore, automating the reduction of chemical mechanisms to speed up the integration of chemical kinetics is an attractive, viable alternative.18,19 A few studies have applied graph theory in reducing mechanisms in atmospheric chemistry, although it is has not widely been used.19–21 Recently, we used a directed-graph path-based automated model reduction (AMORE) approach to develop a novel reduced isoprene oxidation mechanism (version 1.0) which outperformed other common mechanisms in the Framework for 0-D Atmospheric Modeling (F0AM) and Community Multiscale Air Quality (CMAQ) models.19
Here, we incorporate an updated AMORE isoprene oxidation mechanism (version 1.1) into the chemical reaction solver of the GEOS-Chem chemical transport model. We describe the configuration of year-long global and nested-grid eastern US (EUS) simulations. Both ground-based (AQS/IMPROVE) and satellite observations (TROPOMI/CrIS) were used to evaluate the GEOS-Chem model with the AMORE mechanism. We primarily aim to understand the effects of AMORE on GEOS-Chem (1) computational efficiency and (2) accuracy in predictions of air pollutants and related chemical species (PM2.5, O3, NO2, HCHO, and total organic carbon). Finally, we examine the sensitivities of summertime air pollutant levels to biogenic isoprene and anthropogenic emissions across the EUS.
The full isoprene mechanism produces various priority species which are highly sensitive to concentrations of atmospheric reactants. The AMORE algorithm was designed to create highly reduced mechanisms with similar sensitivity to atmospheric reactant concentrations as the full mechanism in regard to priority output products. AMORE reduces the original mechanism to a set of simple mechanistic pathways which are characterized by the types of reactions in the pathway. The type of reaction is defined by the reactive species involved (OH, HO2, NO, NO2, NO3, O3, or methyl peroxy radical) or as a photolytic reaction. AMORE selected the sequence of reaction types necessary in the reduced mechanism to account for sensitivity of the output of the full mechanism to the concentrations of these reactive species as well as photolysis. In order to measure the sensitivity of the output of the full mechanism to these conditions, yields of all priority species were measured under reduced and elevated levels of each of these species (ppb) and photolysis (unitless): OH (1 × 10−6, 1 × 10−4), HO2 (0.04, 0.2), NO (1.17 × 10−6, 0.53), NO2 (1 × 10−4, 0.01), NO3 (2.3 × 10−4, 0.02), O3 (16.7, 100), methyl peroxy radical (0.1, 0.2), and photolysis scale factor (0, 1). Mechanistic pathways consisted of sequences of reaction types and were selected when the species or photolysis involved in the reactions were elevated. For example, the pathway consisting of a reaction with OH followed by a reaction with HO2 was selected when OH and HO2 concentrations were elevated, with all other species or photolysis reduced. The importance of a given pathway was quantified by measuring the change in yield of priority species between the pathway and those with one less step. For example, the yields of priority species in the OH and HO2 two-step pathway were compared to those in the OH one-step pathway and the HO2 one-step pathway, to assess the magnitude of change incurred by the two-step pathway over the one-step pathways.
A set of the most important mechanistic pathways was selected to form a reduced mechanism. In any instances in which two pathways shared common reaction types, the pathways were joined and branched from each other at the differentiating step. Any pathway with more than one step required an intermediate species for each additional step to join the reactions together. For example, the OH and HO2 pathway required an intermediate product for the reaction of isoprene and OH, which then reacted with HO2. All pathways had associated priority species yields, which were used to determine appropriate stoichiometric coefficients for priority species in the terminal reaction of the pathway. Reaction rate constants were selected based on reference values for each reaction type. The algorithmically generated mechanism was run through multiple rounds of box model testing, and manual edits were made to stoichiometric coefficients and rate constants. Some products were included that had been omitted from the algorithm's product selection process as atmospheric reactants. For a detailed description of the development of this AMORE version 1.0 isoprene oxidation mechanism, see Wiser et al. (2023).19
Minor updates were made to the version 1.0 mechanism based on initial GEOS-Chem testing. Most of the isoprene species were renamed to match existing, similar nomenclature accepted in GEOS-Chem. As excessive O3 produced from NO2 in the AMORE version 1.0 mechanism was observed, we opted to remove one artifact, NO2 produced in the isoprene hydroxy hydroperoxide (RIPA) + OH reaction, which lowered O3 to near the baseline levels. In implementing AMORE, carbon and nitrogen are not conserved, and therefore AMORE might not be an appropriate choice for carbon and nitrogen budget studies. Displayed in Tables 1 and 2, the AMORE isoprene version 1.1 mechanism comprises 12 species and 23 reactions. The default GEOS-Chem isoprene oxidation mechanism (mini Caltech) is described in Bates and Jacob (2019)8 and will be referred to as BASE. Compared to BASE, AMORE is smaller by 43 species and 167 reactions. A copy of the BASE mechanism was replaced with our final AMORE version 1.1 mechanism in the Kinetic PreProcessor (KPP) version 2.3.3, the chemical reaction solver of GEOS-Chem.22 Including non-isoprene chemistry, the entire BASE mechanism in KPP consists of 291 species and 903 reactions. KPP takes a list of species, reactions, and rate constants as inputs, solves the chemical kinetics for the given mechanism, and outputs Fortran 90 code that gets compiled in GEOS-Chem.18,23,24 We built compilable code in KPP with the BASE and AMORE mechanisms separately.
# | Species name | AMORE nomenclature | GEOS-Chem nomenclature |
---|---|---|---|
1 | Isoprene | ISOP | ISOP |
2 | Isoprene nitrate peroxy radical | INO2 | INO2B |
3 | Isoprene peroxy radical | IHOO | IHOO1 |
4 | Isoprene hydroxy hydroperoxide | ISHP | RIPA |
5 | Isoprene hydroxy nitrate | IHN | IHN1 |
6 | Isoprene peroxy nitrate | IPN | INPB |
7 | Lumped higher order isoprene nitrates | ISON | IDN |
8 | Isoprene lumped peroxy carbonyl + other functional groups | IPC | IDC |
9 | Isoprene epoxydiols | IEPOX | IEPOXA |
10 | Isoprene-4,1-hydroxyaldehyde | HC5A | HC5A |
11 | Aerosol-phase organic nitrate from isoprene precursors | IONITA | IONITA |
12 | Aerosol-phase IEPOX | SOAIE | SOAIE |
# | Reaction | Rate constant |
---|---|---|
1 | ISOP + O3 → 0.189MVK + 0.58CH2O + 0.25OH + 0.25HO2 + 0.08MO2 + 0.1MCO3 + 0.09H2O2 + 0.461MACR + 0.14CO | 1.58 × 10−14exp(−2000/T) |
2 | ISOP + NO3 → INO2B + 0.3CH2O + 0.3NO2 + 0.3IDN + LISOPNO3 | 2.95 × 10−12exp(−450/T) |
3 | ISOP + OH → IHOO1 + 0.02MO2 + LISOPOH | 2.69 × 10−11exp(390/T) |
4 | IHOO1 + HO2 → RIPA + 0.6HO2 + 0.15CH2O | 4.5 × 10−13exp(1300/T) |
5 | IHOO1 + NO → 0.14IHN1 + 0.7CH2O + 0.44MVK + 0.88HO2 + 0.78NO2 + 0.28MACR + 0.021GLYX | 2.7 × 10−12exp(350/T) |
6 | RIPA + OH → IHOO1 | 4.6 × 10−12exp(200/T) |
7 | INO2B + HO2 → INPB + OH | 3.14 × 10−14exp(580/T) |
8 | INO2B + NO → 0.9CH2O + 0.5MGLY + 0.8MVK + 0.5NO2 + HO2 + 0.2IDN + 0.1MO2 | 9.42 × 10−16exp(580/T) |
9 | INPB + HO2 → 0.8NO2 + 0.4CH2O + 0.05GLYX + 0.1MGLY + 0.4MACR + HO2 + 0.94MVK + 0.2IDN + 0.1MO2 | 3.4 × 10−11exp(390/T) |
10 | IHN1 + OH → IDN + OH + 0.2IEPOXA | 2.4 × 10−7exp(580/T) |
11 | RIPA + OH → 0.15CH2O + 0.05MGLY + 0.15MACR + 0.02GLYX + 0.2MVK + 0.05IDC + 0.58IEPOXA + 0.8OH | 2.97 × 10−11exp(390/T) |
12 | IDC + NO → 0.35NO2 + 0.8NO | 1 × 10−10 |
13 | IDN + OH → CO + 0.12NO2 | 5 × 10−11 |
14 | IDN + NO3 → CO | 2 × 10−14 |
15 | IEPOXA + OH → OH | 5 × 10−11exp(−400/T) |
16 | IHN1 → HNO3 | 2.3 × 10−5 |
17 | IHOO1 + MCO3 → 0.5HO2 + 0.5MO2 + 1.048CH2O + 0.219MACR + 0.305MVK | 8.4 × 10−14exp(221/T) |
18 | IHOO1 + IHOO1 → 2MVK + 2HO2 + 2CH2O | 6.92 × 10−14 [1.1644 + (−7.0485 × 10−4)T] |
19 | IHOO1 + MO2 → MVK + 2HO2 + 2CH2O | 2 × 10−12 [1.1644 + (−7.0485 × 10−4)T] |
20 | IHOO1 + MO2 → CH2O + 0.5HC5A + 1.5HO2 + 0.5MVKHP + 0.5CO + 0.5OH | 2 × 10−12 [−0.1644 + (7.0485 × 10−4)T] |
21 | HC5A + OH → 1.065OH + 0.355CO2 + 0.638CO + 0.355MGLY + 0.283HO2 + 0.125MVKHP + 0.158MCRHP | 4.64 × 10−12exp(650/T) |
22 | Cl + ISOP → HCl + IHOO1 | 7.6 × 10−11exp(500/T) |
23 | RIPA + hν → 0.4CH2O + 0.1MGLY + 0.06MCO3 | PHOTOL(RIPA) |
# | Simulation name | Isoprene chemistry | Emissions perturbation | Horizontal resolution |
---|---|---|---|---|
1 | BASE | Default | None | 2° × 2.5° |
2 | AMORE | AMORE | None | 2° × 2.5° |
3 | BASE_HR | Default | None | 0.25° × 0.3125° |
4 | AMORE_HR | AMORE | None | 0.25° × 0.3125° |
5 | BASE_zUS_ISOP | Default | US biogenic isoprene set to zero | 2° × 2.5° |
6 | AMORE_zUS_ISOP | AMORE | US biogenic isoprene set to zero | 2° × 2.5° |
7 | BASE_zUS_ANTH | Default | US anthropogenic set to zero | 2° × 2.5° |
8 | AMORE_zUS_ANTH | AMORE | US anthropogenic set to zero | 2° × 2.5° |
9 | BASE_zUS_ISOP_ANTH | Default | US biogenic isoprene and anthropogenic set to zero | 2° × 2.5° |
10 | AMORE_zUS_ISOP_ANTH | AMORE | US biogenic isoprene and anthropogenic set to zero | 2° × 2.5° |
11 | BASE_zGLB_ANTH | Default | Global anthropogenic set to zero | 2° × 2.5° |
12 | AMORE_zGLB_ANTH | AMORE | Global anthropogenic set to zero | 2° × 2.5° |
Using boundary conditions from the BASE 2° × 2.5° run, we also performed two 0.25° × 0.3125° nested-grid simulations (BASE_HR and AMORE_HR), over the eastern half of the US: latitude = [24°, 49°] and longitude = [−100°, −66°]. Tracer concentrations are known to be unrealistic on the edges of the nested grid domain where there is no advection.30 Therefore, we applied a buffer zone of 3° along each of the four boundaries. The goal was to focus on peak summertime isoprene emissions over the southeastern US, but we expanded this domain to include the more forested, humid, and densely-populated urban areas east of the 100th west meridian. Particularly in the northeastern US, anthropogenic emissions interact with biogenic emissions to form ground-level O3 in the summertime.31 The higher-resolution model had 47 vertical levels, extending from the surface to 0.01 hPa as in the 72-layer model, but with a coarser vertical resolution in the stratosphere to save computational resources. Transport, cloud convection, PBL mixing, and wet deposition cycled on 5 minutes time steps, while chemistry, emissions, and dry deposition cycled on 10 minutes time steps. Instead of MERRA-2, Goddard Earth Observing System-forward processing (GEOS-FP) meteorology32 was used for the higher-resolution simulations. MERRA-2 and GEOS-FP are similar and both originate from the National Aeronautics and Space Administration (NASA) Global Modeling and Assimilation Office, though GEOS-FP offers the finer native horizontal resolution (0.25° × 0.3125°) that is not available with MERRA-2.
To quantify contributions from biogenic isoprene and anthropogenic emissions to air pollution levels in our EUS domain via both isoprene mechanisms, we perturbed emissions in eight sensitivity simulations (Table 3). Emissions are processed by the Harvard-NASA Emissions Component (HEMCO) module.33 First, we defined a scale factor to zero out emissions over the contiguous US: latitude = [20°, 60°] and longitude [−140°, −50°]. Thereafter, we applied this scale factor to isoprene from the Model of Emissions of Gases and Aerosols from Nature (MEGAN) version 2.1 for the zUS_ISOP simulations.5 Total biogenic isoprene emissions over the EUS domain in the coarser-resolution simulations (12.16 Tg C year−1) and higher-resolution (12.55 Tg C year−1) simulations are comparable. For the zUS_ANTH simulations, the same scale factor was applied to the following anthropogenic emissions sources: the Community Emissions Data System version 2,34 ethane,35 propane,36 the Aviation Emissions Inventory Code,37 ship emissions,34 and anthropogenic fugitive, combustion, and industrial dust.38 These anthropogenic emissions include sulfur dioxide (SO2), carbon monoxide (CO), NOx, PM, and VOCs. Soil NOx emissions including fertilizer NOx are left on and are similar in the coarser-resolution simulations (0.23 Tg N year−1) and higher-resolution simulations (0.22 Tg N year−1). Combining the aforementioned methods, both biogenic isoprene and anthropogenic emissions were zeroed out for the zUS_ISOP_ANTH simulations. Finally, we turned off the same anthropogenic emissions inventories globally for the zGLB_ANTH simulations.
Variable | Units | Number of sites/spatial resolution | Data source |
---|---|---|---|
Surface PM2.5 | μg m−3 | 606 sites | US Environmental Protection Agency AirData45 |
Surface O3 | ppb | 765 sites | US Environmental Protection Agency AirData45 |
Surface total OC | μg m−3 | 43 sites | Colorado State University Federal Land Manager Environmental Database46 |
NO2 tropospheric VCD | molecules cm−2 | 0.05° × 0.05° | NASA Goddard Earth Sciences Data and Information Services Center47 |
HCHO tropospheric VCD | molecules cm−2 | 0.05° × 0.05° | NASA Goddard Earth Sciences Data and Information Services Center48 |
Isoprene total VCD | molecules cm−2 | 0.5° × 0.625° | University of Minnesota Data Repository44 |
The European Space Agency's Sentinel-5 Precursor (S5P) is a sun-synchronous, polar-orbiting satellite with a daily equator overpass time of 13:30 local time.39 TROPOMI is a nadir-viewing spectrometer aboard the S5P, measuring NO2 and HCHO columns in the ultraviolet-visible-near-infrared spectral bands at 3.5 km × 7 km resolution. In addition to higher spatial resolution, TROPOMI has a better signal-to-noise ratio compared with its predecessor, the Ozone Monitoring Instrument (OMI), but still exhibits regional biases including overestimating small HCHO columns and underestimating high HCHO columns.40 We re-gridded TROPOMI Level 2 quality-controlled retrievals, transitioning from reprocessed (RPRO) and offline (OFFL) products on 28 November 2018, to 0.05° × 0.05° daily from June 2018–May 2019. As recommended,41,42 we selected pixels with quality assurance (QA) values greater than 0.75 for NO2 and QA > 0.5 for HCHO, removing errors and observations influenced by cloud, snow, or ice cover. TROPOMI NO2 and HCHO tropospheric VCDs, averaging kernels, and vertical pressure levels were averaged monthly.
CrIS is a Fourier transform spectrometer aboard the US National Oceanic and Atmospheric Administration's Suomi-NPP and NOAA-20 sun-synchronous satellites which have daily overpass times of about 13:30 and 12:40 local time.43 We use the Retrieval of Organics from CrIS Radiances (ROCR) isoprene retrievals44 at 0.5° × 0.625° resolution from Wells et al. (2022)43 who employ a feed-forward neural network to derive total isoprene columns from CrIS hyperspectral range indices. Averaging kernels or other satellite observation operators are not available through this machine learning framework, and the products are averaged each month of year over the period 2012–2020. As a result, GEOS-Chem and CrIS should be compared more qualitatively.
To evaluate GEOS-Chem against TROPOMI and CrIS, we first converted NO2, HCHO, and isoprene mole fractions for each 3D grid box from all GEOS-Chem simulations to partial tropospheric (NO2 and HCHO) or total (isoprene) columns. The calculation required air temperature, specific humidity, surface pressure, and tropopause pressure from both the MERRA-2 and GEOS-FP meteorological products for the coarser and higher-resolution simulations, respectively. For consistency between modeled and observed vertical profiles, we applied TROPOMI averaging kernels at TROPOMI vertical pressure levels nearest to GEOS-Chem vertical pressure levels. This was performed by multiplying partial tropospheric columns (re-gridded to 0.05° × 0.05°) by averaging kernels on the same grid and then integrating from the surface up to the nonuniform tropopause. Isoprene partial columns were simply integrated from the surface to the top of the atmosphere. For all variables listed in Table 4, we selected observations within latitude = [27°, 46°] and longitude = [−97°, −69°] to exclude the buffer zone of the nested-grid simulations. The nearest model grid box to each AQS/IMPROVE site or TROPOMI/CrIS grid box was found, differences between the monthly predicted and observed values were calculated, and these differences were averaged by season and over the year.
Timer name | Horizontal resolution | BASE (hours) | AMORE (hours) | AMORE – BASE time difference (hours) | AMORE – BASE time change (%) |
---|---|---|---|---|---|
GEOS-Chem | 2° × 2.5° | 181.2 | 162.6 | −18.5 | −10.2 |
KPP | 2° × 2.5° | 44.0 | 33.1 | −10.8 | −24.6 |
GEOS-Chem | 0.25° × 0.3125° | 496.3 | 485.9 | −10.4 | −2.1 |
KPP | 0.25° × 0.3125° | 44.9 | 35.8 | −9.1 | −20.3 |
Spatial distributions of the nested-grid EUS simulations compared with each other and against observations are presented in Fig. 2 and 3. On average over the entire time period and at the surface, PM2.5 is lower by 0.08 μg m−3 (1.4%), O3 is lower by 0.70 ppb (2.1%), total OC is lower by 0.02 μg m−3 (1.3%), NO2 is higher by 0.02 ppb (1.7%), and HCHO is lower by 0.17 ppb (9.0%) in AMORE_HR relative to BASE_HR. Fig. 2 shows that PM2.5 is highest in the Great Lakes region (including Pennsylvania) in the model, influenced by anthropogenic emissions (Fig. 2). Both simulations underpredict PM2.5 on average, with the mean model bias just 0.05 μg m−3 (1%) lower in AMORE_HR than BASE_HR. O3 is elevated along the coastline in the model and local areas such as within the Southern Appalachians in the observations. A shallow marine PBL and topography, among other factors, can trap O3.3 Both simulations overpredict O3 on average, but AMORE_HR decreases the mean model bias by 0.55 ppb (1%). Total OC is highest in the southeastern US, and AMORE_HR reduces the mean model bias by 0.02 μg m−3 (1%).
Compared against TROPOMI, both BASE_HR and AMORE_HR slightly overpredict NO2 tropospheric VCD on average to about the same degree, but there are several NO2 hotspots in urban areas that the model underpredicts (Fig. 3). Across the southeastern US, where HCHO and isoprene are enhanced in the domain, both simulations overpredict HCHO tropospheric VCD, but AMORE_HR decreases the EUS mean model bias by 0.17 × 1015 molecules cm−2 (3%). Relative to CrIS, both simulations largely underpredict isoprene total VCD across the EUS domain, possibly because of insufficient isoprene emissions and uncertainties in the satellite retrievals.43 Isoprene is higher by 38.3% in AMORE_HR than BASE_HR at the surface, which reduces the mean model bias across the troposphere by 0.29 × 1015 molecules cm−2 (19%). Higher isoprene domain-wide is beneficial, including for the “isoprene volcano” or Missouri Ozarks region which features a high density of strongly isoprene-emitting oak trees.50 However, AMORE_HR increases the existing high model bias for two prominent isoprene peaks to the south of the Missouri Ozarks.
Higher isoprene in AMORE_HR than BASE_HR suggests that less of it is oxidized in the AMORE mechanism. The main oxidant, OH, exhibits widespread lower annual mean concentrations, by 3.4% at the surface and by 1.2% across the troposphere on average globally, with the AMORE mechanism than BASE mechanism (see Fig. S1 in the ESI†). In low-NOx conditions, reactions between isoprene peroxy radicals (IHOO1) and HO2 leads to OH consumption via hydroperoxide (RIPA) production which can be regenerated through more intramolecular hydrogen shifts.8 Less OH recycling and thus lower OH concentrations are expected with fewer intermediate species in the AMORE mechanism, on par with other similarly sized reduced mechanisms.19 Higher isoprene may also emphasize the low-NOx pathway, leading to further OH depletion and hindering NO2 production. Reaction rates for ISOP + OH and ISOP + O3 are the same order of magnitude but lower and higher in the AMORE mechanism, respectively. Since the rate constants are about the same in both mechanisms, changes in species concentrations are likely responsible for the reaction rate differences. While the latter reaction depletes O3 and is more competitive in the AMORE mechanism, ISOP + OH happens faster, and lower HCHO signifies the impact of less OH-initiated oxidation. A lower ISOP + NO3 reaction rate in the AMORE mechanism than BASE mechanism, due to a rate constant discrepancy, could allow less NOx to be lost to isoprene nitrates. We find that the ISOP + NO3 rate constant in the AMORE mechanism matches that in Wennberg et al. (2018),17 lower by 1–2 orders of magnitude than in the BASE mechanism.
As displayed in Fig. S1 in the ESI,† global spatial differences in annual mean surface concentrations of PM2.5, O3, and other key chemical species between the BASE and AMORE simulations are mostly small and have the same signs as over the EUS. Isoprene chemistry most prominently differs between the two mechanisms over NOx-limited tropical forest regions, namely the Amazon, Central Africa, and Indonesia, where isoprene is emitted in significant amounts and exhibits proportional differences. While PM2.5, biogenic OA, and IEPOX OA are lower in these tropical forest regions with the AMORE mechanism, IEPOX OA is higher elsewhere, revealing the impact of missing non-IEPOX pathways on SOA formation. Comparing the two mechanisms for several other species of broad interest, we show that annual mean surface CO and peroxyacetyl nitrate (PAN) concentrations are lower by 2.5% and 3.1%, respectively, on average globally in AMORE. For NO3, the sign of the difference varies from region to region, with NO3 clearly lower in AMORE across India. Nitric acid (HNO3) also varies regionally, but it appears to be lower in AMORE over parts of Asia and Africa where HNO3 concentrations are relatively high.
Although higher isoprene with the AMORE mechanism reduces the global mean model bias (Fig. S2†), as background isoprene levels are underestimated, it enhances the existing overestimation for isoprene hotspots, particularly the Amazon region. Again, due to less oxidation by OH, there are widespread lower surface HCHO concentrations in AMORE, decreasing the existing positive tropospheric HCHO bias across these tropical forest regions. Globally, surface NO2 levels are slightly higher in AMORE than BASE, especially over India, China, and other high-NOx areas. We find that GEOS-Chem underestimates NO2 tropospheric VCD over Europe, which may be due to other factors such as emissions, and that NO2 in AMORE is in the right direction in such regions. Depending on the specific application of the model, the BASE mechanism might be a better choice over tropical forest regions to capture detailed chemistry, whereas the AMORE mechanism may suffice for estimating ambient air pollutants over much of the globe. However, more measurements of chemical species, including radical species such as OH, are needed to further evaluate the GEOS-Chem model.
Unlike PM2.5 and O3, surface HCHO and total OC are much more sensitive to isoprene, with increases of 49–54% (HCHO) and 51% (total OC) from US biogenic isoprene emissions alone and increases of 68–71% (HCHO) and 64% (total OC) from the combination of US isoprene and anthropogenic emissions. US anthropogenic emissions are responsible for the majority of surface NO2 (84–85%) which decreases slightly when adding in biogenic isoprene emissions. For all air pollutants, little differences exist between the BASE and AMORE simulations, suggesting that we can use the computationally-efficient AMORE approach without sacrificing process-level understanding gained from the perturbation simulations.
Future development of reduced chemical mechanisms can benefit from our automated graph theory-based approach. We tested the AMORE algorithm for gas-phase isoprene oxidation, but there are potential applications to aqueous-phase chemistry or other mechanisms. More work is needed to make AMORE fully automated and sufficiently flexible for use in a wide range of chemical mechanisms and atmospheric models. A limitation of this study was the amount and quality of observational data available to evaluate the GEOS-Chem model. This model evaluation would benefit from more in situ and remote sensing measurements of chemical species important to isoprene oxidation, such as OH, HCHO, and OA, in different photochemical regimes across the globe. For example, the Tropospheric Emissions Monitoring of Pollution (TEMPO) instrument, launched in April 2023, will provide daytime hourly air pollution information across North America from a geostationary orbit.52 We are interfacing with the GEOS-Chem modeling team to make AMORE an option for all users in the standardized GEOS-Chem code. Improvements in computational speed and accuracy of CTMs and CCMs will allow for efficient air quality and climate forecasting, research, and management.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3ea00121k |
This journal is © The Royal Society of Chemistry 2023 |