Air pollution can significantly impact air quality (Akimoto, 2003; Fiore et al., 2002; Jacob et al., 1993), human health (Anenberg et al., 2019; Huang et al., 2020; Lelieveld et al., 2015), and climate change (Monks et al., 2015; Ramanathan et al., 2002; Ramanathan and Feng, 2008; Unger et al., 2010). Although air quality in the United States has substantially improved since the implementation of the Clean Air Act of 1990, tropospheric ozone (O₃) still poses a challenge to many regions in the United States (Cooper et al., 2014, 2015; Jaffe and Ray, 2007). O₃, a secondary air pollutant formed through the photochemical interactions between its gas-phase precursors, i.e., nitrogen oxides (NOx= NO + NO₂) and volatile organic compounds (VOCs), often exceeds allowable limits for O₃ established by the National Ambient Air Quality Standards (NAAQS) set by the United States Environmental Protection Agency (US EPA) (i.e., a maximum daily 8 h average (MDA8) of 70 ppbv or less) in various US cities, despite significant reductions in its precursor species.

Southeast Michigan (SEMI) has often been classified as a non-attainment area for O₃ (US Environmental Protection Agency, 2021). SEMI has experienced historically high levels of air pollution from being heavily concentrated with industry (e.g., coal-fired power plants, steel and cement facilities, petroleum refineries, and incinerators) and is subject to various mobile emissions sources due to its proximity to highways and the US–Canada ports-of-entry (in Detroit and Port Huron). Elevated O₃ levels have been associated with a variety of negative impacts to human health, agriculture, and the natural environment, which include premature deaths attributable to respiratory and cardiovascular diseases (Sicard et al., 2018), impacts to crop yields resulting from reduced photosynthesis (Fuhrer et al., 1997; Wortman and Lovell, 2013), and reduced visibility due to photochemical smog. The Michigan–Ontario Ozone Source Experiment (MOOSE) (Olaguer et al., 2023) was carried out to define potential O₃ attainment strategies in SEMI and better understand what contributes to O₃ exceedances in the region. It was a multi-institution (e.g., Michigan Department of Environment, Great Lakes, and Energy (EGLE), Environment Climate Change Canada (ECCC), Ontario Ministry of Environment, Conservation, and Parks (MECP), and university partners) campaign that was carried out in two phases: Phase I (24 May to 30 June 2021) and Phase II (6–28 June 2022). The MOOSE observations included a mobile lab with detailed measurements of ozone and its precursors, ground-based remote sensors (i.e., Pandora), and an airborne remote sensor (i.e., GEO-CAPE Airborne Simulator). Previous studies in Michigan have mainly investigated the impact of lake breezes on air quality (Abdi-Oskouei et al., 2020; Acdan et al., 2023; Brook et al., 2013; Dye et al., 1995; Hanna and Chang, 1995; Stanier et al., 2021; Vermeuel et al., 2019) and the connections between human health adversities and air pollution (Cassidy-Bushrow et al., 2020; Lemke et al., 2014), with little attention focused on O₃ atmospheric chemistry in SEMI. Xiong et al (2023) was the first to use a combination of MOOSE campaign measurements in 2021 and box modeling to investigate O₃ formation regimes in SEMI and found that summertime O₃ is limited by VOC emissions but pointed to uncertainties due to the small number of days used for the analysis. Because the spatial distribution of O₃ is dependent on precursor emissions, location, and meteorology, local O₃ production and loss in SEMI may be largely different compared to other regions.

Models provide credible, process-based mathematical representations of chemistry–climate interactions in the atmosphere (Brasseur and Jacob, 2017). O₃ biases have been identified in various global chemistry–climate models, with suggestions for improvements, such as better representation in temperature, anthropogenic emission inventories, and deposition (Schwantes et al., 2022). However, in many of these cases, the global models were not being run at horizontal and vertical resolutions fine enough to simulate ozone production and loss accurately (Schwantes et al., 2022). Although current models are efficient in reproducing rural pollutant concentrations, surface O₃ bias persists, which can be attributable to a coarse (> 100 km) grid's ineffectiveness at reproducing urban sources and transport (Jo et al., 2013; Monks et al., 2015). The large grid cells in coarse grids artificially dilute local emissions of O₃ precursors, imported pollution plumes, and topography, which can alter abundance and mixing at the surface (Monks et al., 2015). There have been advancements in the use of high horizontal grid resolutions (1–28 km), which have the potential to produce more realistic simulations of O₃ production and loss. MUSICA (Multi-Scale Infrastructure for Chemistry and Aerosols) is a state-of-the-science unified modeling framework, allowing for seamless global and regional simulation within one model with consistent dynamics and chemistry (Pfister et al., 2020). The initial implementation of MUSICA (MUSICAv0) is a configuration of CAM-chem (the Community Atmosphere Model with chemistry) available in the Community Earth System Model Version 2 (CESM2; Danabasoglu et al., 2020) using the spectral element (SE) dynamical core, allowing for regional refinement. Several studies have taken advantage of MUSICAv0's regional refinement capabilities using custom grid applications. Schwantes et al. (2022) evaluated horizontal resolution and chemistry at varying scales (∼ 111 and ∼ 14 km) over the Southeastern US and found that O₃ was better simulated over urban regions, particularly using the ∼ 14 km grid and updated isoprene and terpene chemistry. Tang et al. (2022) included plume rise and a diurnal cycle of fire emissions in MUSICAv0, using the standard ∼ 14 km resolution over the contiguous US (CONUS), and found that this addition improved MUSICAv0 simulations compared with observations. Tang et al. (2023) developed a custom grid over Africa at ∼ 28 km in MUSICAv0, compared it to the regional model WRF-Chem (Weather Research Forecast with Chemistry), and found that MUSICAv0 performance was comparable to that of WRF-Chem when comparing to satellite and surface measurements of O₃ and carbon monoxide (CO). Jo et al. (2023) constructed two global (∼ 112, ∼ 56 km) and two regional (∼ 14, ∼ 7 km) refinement grids over South Korea for use in MUSICAv0 and found that grid resolution can heavily impact model evaluation near the surface, in particular within urban regions, as well as strongly affect the oxidation of VOCs. Lichtig et al. (2024) used a custom grid over South America with a resolution of ∼ 28 km to quantify the local and long-range origins of CO in the region. Edwards et al. (2024) used MUSICAv0, along with the Geostationary Environment Monitoring Spectrometer (GEMS), to study NO_x over Northeast Asia and Seoul, South Korea, to distinguish different emission sources. As can be noted from previous work, custom grids have been used to understand an extensive range of atmospheric physical and chemical processes.

In this study, we created a custom regional refinement grid over the state of Michigan in the United States with a horizontal resolution of 1/16° (∼ 7 km) and compared it to the standard MUSICAv0 1/8° (∼ 14 km) grid over CONUS. We used the Community Mesh Generation Toolkit, which is available to the community and provides the necessary tools for defining a high-resolution grid mesh (i.e., generating input files). A sector-based diurnal cycle was applied to anthropogenic nitric oxide (NO) emissions and was included for each resolution (Crippa et al., 2018; Jo et al., 2023). In total, four simulations were run during Phase I of the MOOSE campaign, which included a variety of high-resolution measurements used for model evaluation. This work focuses on evaluating the model simulations with measurements from MOOSE, the differences between the regional refinement grids, and changes that result from the application of the diurnal cycle for anthropogenic NO emissions.

In this section, we evaluate MUSICAv0 model results with observations from the MOOSE field campaign in May to June 2021. For evaluation, we compare the models with O₃ and NO₂ from MI EGLE stationary sites, a range of gas-phase species from the AML, NO₂ and HCHO columns from two Pandora spectrometers in SEMI, and NO₂ columns from GCAS. We evaluate the models using diverse datasets from MOOSE for a comprehensive analysis, as no single dataset has the ability to capture all aspects of atmospheric composition (e.g., emissions, chemistry, transport, meteorology). These different datasets can also help capture different aspects of a model such as near-surface chemistry (i.e., in situ measurements) and column burdens (i.e., aircraft-based remote sensing) to determine model skill, characterize model errors, improve model representation, and measure our confidence in the model results for reproducing reality.

For the comparison, we match the observed mixing ratios to the closest model grid point at each time. Modeled NO₂ and HCHO columns are calculated for the troposphere using the NO₂ and HCHO mixing ratios at each level of the model and multiplying them by the number density of air, which changes with altitude due to decreasing pressure, to get the number concentrations. Once the number concentrations are obtained, we multiply them by the layer thickness and integrate up to the average height of the column (i.e., for Pandora, the approximate height used was ∼ 3 km; for GCAS, the altitude of the aircraft was ∼ 12 km).

O₃ concentrations are highly associated with NO₂, where NO_x, in general, plays a critical role in the photochemical production and destruction of O₃ in the presence of sunlight. O₃ production in the troposphere is largely dependent on the availability of NO_x and VOCs and can give great insight into O₃ control. This dependency is classified into NO_x- and VOC-limited regimes. In a NO_x-limited regime, the rate of O₃ production relies on the abundance of NO_x and increases with NO_x concentrations but is not dependent on the concentrations of VOCs (Wang et al., 2019). In action, decreasing NO_x concentrations would lead to reductions in O₃ (Jacob, 1999). On the other hand, in a VOC-limited regime (or NO_x-saturated regime), the rate of O₃ production increases with VOC concentrations and is not dependent of NO_x (Wang et al., 2019); therefore, reducing the amount of VOCs would lead to reductions in O₃ (Jacob, 1999). The chemical relationship among O₃-NO_x-VOCs is critically important for defining mitigation strategies set to improve O₃ from region to region.

The previous section (Sect. 3) evaluated four MUSICAv0 simulations using two different regional refinement grids (ne30x8, ne30x16) and the addition of a diurnal cycle for anthropogenic NO emissions (ne30x8 DIUR, ne30x16 DIUR) against observations from Phase I of the MOOSE field campaign. Building on the evaluation in Sect. 3, this section discusses the differences resulting from changes in horizontal grid resolution and the application of the diurnal cycle for anthropogenic NO. We analyze how site-specific behaviors are driven by the model changes and how those behaviors drive O₃ formation. First, the spatial distributions in the two resolutions are compared (Figs. 12–14), and then the diurnal cycles in different environments are compared (Figs. 15–18).

Figure 12 shows emissions of NO across the SEMI region and the different grid boxes pertaining to the (a) ne30x8 and (b) ne30x16 horizontal resolutions. The Allen Park and Trinity-St. Mark's ground sites (black triangle and diamond, respectively) are also shown relative to their grid box locations and the distribution of NO emissions. This figure shows that in the coarse resolution (Fig. 12a), the two sites (a suburban and an urban site) are represented by the same grid box, whereas in the finer, ne30x16 resolution (Fig. 12b), they are present in distinct grid boxes. Although the total emissions for a region are the same, emission fluxes can become more resolved when moving to a finer grid resolution (Jo et al., 2023), which will ultimately impact model simulation evaluation as the horizontal resolution becomes finer and finer.

To quantitatively assess the impact of the finer resolution on the simulation of ozone and its precursors, the ne30x16 (7 km) results have been conservatively regridded to the ne30x8 grid. These regridded results illustrate the impact model resolution can have on atmospheric chemistry, depending on the compound. Figure 13 shows the modeled monthly averaged carbon monoxide (CO) concentrations at the surface for June 2021 for the ne30x8 (Fig. 13a), ne30x16 (Fig. 13b), and conservatively regridded model output from the ne30x16 simulations to the ne30x8 horizontal resolutions (Fig. 13c), respectively. CO is mainly emitted through incomplete combustion processes and has a generally long lifetime, lasting from week to months, allowing it to be transported over long distances (Gaubert et al., 2016). These characteristics make CO relatively chemically inactive. Because of this, there are minimal chemistry effects, where the majority of impacts on CO will result from the grid resolution. Fine-scale features of CO are better captured in the ne30x16 horizontal resolution simulations, as CO concentrations are more resolved over urban regions, as can be seen in Fig. 13b. Figure 13c shows the modeled CO concentrations, conservatively regridded from the ne30x16 horizontal resolution to the ne30x8 horizontal resolution. Using this regridding method to go from the finer to the coarser resolution did not reproduce the same results seen from running the ne30x8 simulation (Fig. 13a). When the model is run at 1/16° horizontal resolution, localized features (e.g., pollution plumes, sharp emission gradients) are better resolved, and land use is better represented. Figure 13a–c have mean CO concentrations over SEMI of 141.7, 131.1, and 132.1 ppb, respectively.

On the other hand, O₃ is highly chemically active. Figure 14 shows modeled monthly averaged O₃ concentrations at the surface for June 2021 for ne30x8 (Fig. 14a) and ne30x16 (Fig. 14b), as well as the conservatively regridded model output from the ne30x16 simulations to the ne30x8 horizontal resolutions (Fig. 14c), respectively. Similarly to what Jo et al. (2023) found over South Korea, there is a decrease in O₃ concentrations over urban areas in SEMI as a result of NO_x titration. This reduction in O₃ is more prominent with the finer horizontal model grid resolution, which leads to differences in the monthly mean surface O₃ concentrations in coarse (40.1 ppb) and fine (38.0 ppb) horizontal resolutions over SEMI. When regridding O₃ concentrations from the finer to the coarser horizontal resolution, the NO_x titration is visible over the urban areas in SEMI, similarly to the ne30x16 simulation, but it is stronger when compared to the ne30x8 simulation.

Figure 15 shows the diurnal variation for O₃, NO, and NO₂ concentrations, NO emission flux, and the planetary boundary layer height from the four simulations presented in this study for three sites – a suburban downwind site (Allen Park), an urban site (Trinity-St. Mark's), and a rural site (New Haven). The Allen Park and Trinity-St. Mark's sites are located within the same grid box in the ne30x8 simulations, while in the ne30x16 simulations, they are not. Figure 15a–c show that horizontal resolution had the most impact on O₃ concentrations at all sites during peak times (12:00–18:00 EDT), with differences between simulations of up to ∼ 5 ppb. This difference results in an improvement for the ne30x16 simulations based on the findings in Fig. 4, where peak O₃ performed best in the finer-resolution simulations when compared to the surface sites. The addition of a diurnal cycle for anthropogenic NO did not have a significant impact on O₃ concentrations during these peak times but saw larger differences during the 05:00–11:00 EDT time frame, likely as a result of lower NO (Fig. 15d–f) and NO₂ (Fig. 15g–i) concentrations and associated NO_x titration in the model simulations. It is important that we acknowledge the differences caused in NO and NO₂ at the Allen Park and Trinity-St. Mark's sites due to the grid resolution. As was mentioned before, Allen Park and Trinity-St. Mark's are located within the same grid box in the coarser resolution. When using the ne30x16 horizontal resolution, higher concentrations for NO and NO₂ can be seen at the Trinity-St. Mark's site than at the Allen Park site, which coincides with the high urbanization in that area. Although these differences are not greatly significant to O₃ concentrations, it is important that urban and non-urban areas are distinguished, as they can have higher emissions fluxes (Fig. 15j–l).

On the other hand, a more rural site, like New Haven, has just as much O₃ as an urban site, even though NO emission fluxes are quite low. This is likely a result of the New Haven site being more representative of background concentrations driven by transport from upwind areas that include close proximity to a major highway and near Lake St. Clair. Hayden et al. (2011) found that along the Lake St. Clair shore, pollutants can be confined, leading to elevated pollutant concentrations and an increase in oxidizing capacity.

Biogenic VOC emissions can be heavily impacted by changes in model horizontal grid resolution, as they are based on meteorological parameters, such as temperature, because they are calculated online in the land model using the MEGANv2.1 algorithm (see Sect. 2.1.3). Figure 16 shows isoprene mixing ratios averaged over June and July 2021 from the ne30x8 (Fig. 16a) and ne30x16 (Fig. 16b) simulations, where the ne30x8 simulation shows about double the amount of isoprene compared to the ne30x16 simulation spread over a wider area. This can be explained by the higher isoprene emission fluxes in SEMI in the coarse resolution (Fig. 17a) compared to the finer resolution (Fig. 17b). These differences in isoprene emissions caused by the different horizontal resolutions directly impact isoprene concentrations in the models. These findings are directly supported by Fig. 3, where although temperatures between the simulations are not significantly different, there are changes in cloud totals and winds that could impact solar radiation and thus the isoprene emissions. The differences in temperature between the resolutions are also illustrated in the maps in Figs. S16–S31 in the Supplement.

Similarly to O₃ and NO₂, isoprene has a strong diurnal cycle that is driven by temperature and solar radiation. The diurnal cycles for isoprene, HCHO, and the hydroxyl radical (OH) are shown in Fig. 18 for the same three sites in Fig. 15. For the Allen Park and Trinity-St. Mark's sites, the isoprene mixing ratios (shown in Fig. 18a–c) were generally lower in both simulations, which coincides with suburban and urban landscapes that have relatively low densities of trees, while at the New Haven Site, the concentrations were about double compared to the other two sites, as it is a more rural region. For all of the sites, the isoprene concentrations were lower in the ne30x16 simulations compared to the ne30x8 simulations. In the ne30x8 simulations, the isoprene concentrations at the Allen Park and Trinity-St. Mark's sites are shown to be the same, but when using the ne30x16 resolution, isoprene is shown to be lower in the urban location compared to the suburban location. This indicates that finer resolution can help better characterize regions of interest and assist in the misclassification of emission sources, which coincides with the findings in Sect. 3. The lower isoprene concentrations also impact HCHO concentrations (shown in Fig. 18d–f), as HCHO is a product of isoprene oxidation. HCHO is lower in the ne30x16 simulations, which coincides with the lower isoprene concentrations. OH concentrations (shown in Fig. 18g–i) are generally consistent in all simulations, with very minimal changes after applying the diurnal cycle for anthropogenic NO. Because SEMI is in a more VOC-limited regime (Xiong et al., 2023), OH concentrations are less sensitive to changes in VOCs and more prone to changes resulting from the changing NO_x levels due to titration of O₃ (de Gouw et al., 2019). HCHO is NO_x sensitive, meaning that more HCHO is produced in the presence of higher NO_x concentrations (Schwantes et al., 2022). HCHO was not heavily changed by the application of the diurnal cycle for anthropogenic NO emissions, indicating that the main driver for changes in HCHO is grid resolution. In general, isoprene and HCHO decrease with increased resolution, while OH remains relatively constant between model simulations.

The findings of this study show that O₃ production in SEMI is strongly governed by the spatial distribution of emissions and different chemical regimes. The urban location analysis showed that Detroit, which is a major industrial hub in the region, is consistent with a VOC-limited regime, where O₃ concentrations are suppressed by high NO_x titration in the daytime but can become sensitive to changes in VOCs during peak O₃ times. The suburban and remote location analysis (i.e., Allen Park and New Haven, respectively) showed that they are in a more NO_x-limited regime, where higher BVOCs and lower NO_x titration can lead to more efficient O₃ production. The spatial distribution is seen more clearly as we move towards finer resolutions, indicating more realistic emissions.

In VOC-limited regimes, targeting reductions in VOCs, such as those from the industrial sectors, is crucial compared to reductions in NO_x, as it could lead to temporary increases in O₃ production. In NO_x-limited regimes, where NO₂ drives O₃ production, reductions in transportation emissions and long-range transport would decrease O₃. The improvement in model representation of NO₂ and, in turn, O₃ during rush hour times (Figs. 4–5) shows how emissions can be misrepresented in the models. It is necessary that future work considers incorporating higher-resolution temporal profile and regional emissions to better distinguish different O₃ processes. Future work should also explore the impacts of targeting the contribution of different emission scenarios in SEMI to demonstrate the impact of different regulatory decision-making outcomes.

Tropospheric O₃ in SEMI is a persistent problem in the region, majorly resulting from anthropogenic activity. MUSICAv0, a global chemistry–climate model with regional refinement capabilities, has allowed us to evaluate whether the MUSICA framework is suitable for studying O₃ atmospheric chemistry in local, urban environments. This study aimed to evaluate the impact of horizontal grid resolution and diurnal cycles (for anthropogenic NO emission) on MUSICAv0 simulations, using a custom grid over the state of Michigan with a resolution of 1/16° and leveraging a suite of measurements from Phase I of the MOOSE field campaign in 2021.

For O₃ and its precursors, both grid resolution and the diurnal cycle for anthropogenic NO emissions were important, largely depending on the time of day and the region. Horizontal grid resolution was important for O₃ during peak O₃ times (12:00–18:00 EDT), but during the night and early morning, O₃ was largely impacted by the application of the diurnal cycle as a result of changing NO₂ during NO_x peak times.

This work compares simulated NO₂ and HCHO tropospheric columns from MUSICAv0 model runs to measurements from the Pandora network at two sites in SEMI for the first time. The NO₂ columns from Pandora agreed with the temporal variability of the NO₂ columns at the two urban sites, where the application of the diurnal cycle for anthropogenic NO emissions at both resolutions generally made the model perform better during peak NO₂ times but resulted in greater model overestimations in the later part of the day. These trends are important because they can be indicative of high anthropogenic emission sources from industry and transportation in the model. Modeled HCHO columns compared to Pandora, on the other hand, were largely impacted by a combination of grid resolution and the diurnal cycle, where grid resolution impacts HCHO because of online calculations of biogenic VOCs and changing NO_x levels can promote VOC oxidation, leading to lower HCHO columns in the model. These changes led to underestimations of HCHO tropospheric columns in the model compared to the observations. This underestimation indicates that the model simulations are not capturing anthropogenic VOCs efficiently.

In addition,the NO₂ tropospheric columns from the model simulations were compared to observations from GCAS for the first time. This comparison showed that the finer resolution captured more pronounced pollution plumes corresponding to observed wind directions, which can be important when assessing transport from more localized sources. As the grid resolution in global chemistry–climate models becomes finer, future work should consider using NO₂ column data from remote sensing instruments to develop regional emission inventories for more fine-scale applications.

This work showed that grid resolution is more important for O₃ precursors (i.e., NO_x, HCHO, isoprene) than for O₃ itself, which agrees with the findings in Jo et al. (2023) and Schwantes et al. (2022). Changes due to grid resolution were largely a result of the artificial mixing of emissions. Finer resolutions can better classify source regions and distinguish between urban and non-urban regions. Grid resolution also impacted biogenic VOCs, as they are calculated online via MEGANv2.1 based on various meteorological parameters. Although isoprene in the finer-resolution simulations showed better performance compared to the AML measurements, SEMI is generally not prone to high isoprene emissions. Future work using the regional refinement grid over Michigan should focus on evaluating locations with higher vegetation density.

Applying diurnal cycles for anthropogenic NO on monthly emissions also played a crucial role in nighttime O₃ chemistry. The diurnal cycle often impacted O₃ and precursor concentrations more than the grid resolution. Future work should evaluate the impacts of applying diurnal cycles to more anthropogenic emissions other than NO. In addition, we acknowledge that apart from applying a diurnal cycle for anthropogenic NO emissions, the evolution of the PBL can also play a significant role in the formation of O₃ and NO_x. In the daytime, a rising PBL can mix surface NO_x and VOCs upwards, reducing O₃ concentrations near the surface, while in the nighttime, a shallower PBL can trap emissions near the surface, leading to higher NO_x titration. Uncertainties associated with the PBL could lead to underpredictions of NO₂ in the model and misrepresentations of O₃ peaks.

This is one of a few studies evaluating O₃ production and loss processes with custom grids in MUSICAv0. Although O₃ biases still persist in the MUSICAv0 simulations over this region, these biases are generally lessened with finer grid resolution during peak O₃ times and with diurnal cycles for anthropogenic NO during the nighttime. This case study is limited to SEMI, which can have different implications compared to previous work. For example, the state of Michigan is about 2.5 times larger than South Korea, which was studied in Jo et al. (2023) using a similar methodology, and has a completely different topography. Michigan is generally flat and surrounded by freshwater lakes, as opposed to South Korea's mountainous terrain surrounded by ocean encompassing a megacity. Schwantes et al. (2022) found that O₃ was better simulated over urban regions across the Southeastern US, especially when using a ∼ 14 km regional refinement grid and updated chemistry in MUSICAv0. This work took into consideration a finer grid resolution mesh (∼ 7 km) and compared to ∼ 14 km to show that regional refinement improves O₃ representativeness in the model. Future work aims to take advantage of custom grids to quantify the contribution of emissions and transport to O₃ atmospheric chemistry in the region. Future work should also take into consideration the use of a more updated version of the CAMS-GLOB-ANT emissions, as well as the diurnal variation profiles of CAMS-GLOB-TEMPO (Guevara et al., 2021; Soulie et al., 2024), or more regional emission inventories such as the National Emission Inventory (NEI) from the US EPA. Optimization of a regionally refined, coupled model such as MUSICAv0, through resolution and emission modeling studies, can have significant implications for the design and development of effective surface O₃ mitigation strategies in SEMI.