In box-model tests against 16 atmospheric chamber experiments, the original mechanism systematically underestimated aerosol concentrations, whereas the expanded schemes were generally within a factor of two and usually within 30% of observations. The study focused on benzene, the simplest aromatic compound, and phenol, the main product of benzene oxidation by the hydroxyl radical. Benzene is emitted by traffic, cigarette smoke, and industrial solvent use, including applications in textiles, pesticides, and pharmaceuticals.

Phenol is also emitted directly by biomass burning, traffic exhaust, and manure spreading in agriculture. Oxidation can transform these volatile organic compounds into more functionalized, lower-volatility products that enter the particle phase and form SOA. Existing three-dimensional air-quality models commonly parameterize this process with yields derived from atmospheric simulation chambers or with a limited set of representative surrogate compounds.

Although computationally efficient, those approaches do not explicitly resolve the molecular composition of the particle phase or the full network of reactions responsible for producing it. The team began with the quasi-explicit Master Chemical Mechanism, or MCM, which represented benzene degradation and first-generation oxidation products but omitted several pathways supported by recent experiments and theoretical calculations. The expanded mechanism permits repeated hydroxyl addition to phenolic rings, producing trihydroxybenzene, tetrahydroxybenzene, and pentahydroxybenzene in addition to the previously represented phenol and catechol.

It also incorporates autoxidation sequences for peroxy and alkoxy radicals, in which successive intramolecular reactions are followed by oxygen addition. Termination of these radical sequences can produce hydroperoxides, organonitrates, carbonyl compounds, and very low-volatility peroxide dimers formed by accretion between peroxy radicals. A separate pathway represents cyclic epoxides produced when hydroxyl oxidation of benzene-derived bicyclic peroxides and nitrates competes with fragmentation and other oxidation routes.

Several baseline assumptions were revised in light of more recent evidence. First-generation production of an open-chain epoxide was removed because experimental work indicates a very low yield, while the branching ratio leading to formation of a benzene-derived bicyclic peroxy radical was increased from 35.2% to 47%. The corresponding bicyclic alkoxy radical was also allowed to compete between decomposition and a hydrogen-shift pathway.

The latter produces a ketene–enol intermediate that rapidly oxidizes to formyl ketene, glyoxal, and formic acid, and an additional route accounts for acrolein formation. Because most end products introduced through the new pathways have low saturation vapor pressures, they partition almost entirely into the particulate phase. Loss reactions were retained instead for compounds with sufficiently high volatility to occur in both gas and particulate phases, using estimated reaction rates for photolysis and reactions with hydroxyl, nitric oxide, and nitrate radicals.

Particular attention was given to bicyclic peroxy radicals formed when hydroxyl addition occurs at a carbon already carrying a hydroxyl group, creating a geminal-diol structure. Such intermediates can rearrange rapidly and yield highly oxygenated organic molecules, including C6 carboxylic acids and, in the case of catechol, diacids. Because this minor reaction route can make a disproportionately large contribution to SOA mass, the relevant branching ratios and rate coefficients are important.

The researchers applied multiconformer transition-state theory, which incorporates multiple low-energy conformers of reactants and transition states, together with an Eckart treatment of quantum tunnelling. Calculations were performed at several density-functional and correlated levels of theory. A correction previously found to generate artefacts for hydrogen shifts from hydroxyl groups was deliberately not applied to that reaction class.

The calculations showed that endocyclization, which creates the five-membered oxygen bridge needed to form a geminal-diol bicyclic peroxy radical, competes with hydrogen shifts and, for catechol, with hydroperoxyl loss. The resulting endocyclization branching ratios were 10% for phenol and 59% for catechol. Because catechol can add oxygen at two different sites and quantum chemistry could not reliably resolve the relative probabilities, the researchers tested oxygen-addition branching ratios of 10% and 50% for the route leading toward endocyclization.

They also evaluated two possible ways of breaking a carbon–carbon bond during molecular rearrangement. Rearrangement of the phenol-derived radical could generate products containing eight oxygen atoms, while one catechol-derived route could proceed to products with as many as 11 oxygen atoms. Alternative rearrangements arising from hydroxyl addition at an ortho position were calculated to be too slow to compete under atmospheric conditions and were not included in the model.

Evaluation used version 2.0 of the SSH-aerosol multiphase box model, which represents gas-phase chemistry, aerosol dynamics, and partitioning among gas, organic, and aqueous phases. The simulation set covered eight phenol and eight benzene chamber experiments conducted over eight hours under natural daylight, with radical production driven by photolysis. The experiments varied nitrogen oxide levels and included conditions with no seed aerosol, acidic inorganic seeds, or neutral seeds.

Initial precursor concentrations ranged from 79 to 264 parts per billion for phenol and from 168 to 325 parts per billion for benzene, substantially above ambient concentrations so that oxidation and SOA formation could be measured over the chamber timescale. The original MCM produced SOA concentrations far below observations. Adding the newly implemented pathways greatly increased simulated particle mass, with autoxidation particularly influential at low nitrogen oxide levels and successive hydroxylation becoming more prominent under high-nitrogen-oxide conditions.

The addition of geminal-diol chemistry generally improved agreement with the chamber measurements, although it sometimes reduced mass that the preceding mechanism had overestimated. Simulations for phenol with neutral seeds remained as much as 1.9 times below observations. The researchers identified uncertainty in the gas–particle partitioning of intermediate-volatility carbonyls and quinones as one possible contributor, along with aqueous-phase hydration of light aldehydes when seed concentrations were high.

Sensitivity tests also showed that assumptions about the catechol cyclization branch can substantially alter the results: reducing the favored branching ratio from 50% to 10% lowered simulated SOA mass in most experiments and increased disagreement with observations. Across the experiments, the expanded mechanism had a mean relative error of 28.2%, compared with 83.2% for the original MCM. To make the detailed mechanism usable in three-dimensional air-quality models, the researchers reduced it with GENOA, an automated framework that removes, jumps, or lumps chemical species while retaining pathways important for SOA formation.

Across 9,433 European atmospheric conditions, the final scheme contains 30 reactions and 28 species, including 12 condensable compounds, compared with 1,772 reactions and 1,160 species in the detailed mechanism. It preserves successive hydroxylation, peroxy- and alkoxy-radical autoxidation, and geminal-diol bicyclic-peroxy-radical rearrangements. Its mean SOA-mass error relative to the expanded reference was 3.94%, with a maximum error of 22.5%; errors remained below 5% across most locations.

The study notes that intraparticle chemistry remains an important source of uncertainty, particularly for phenolic compounds that may undergo photochemical reactions within biomass-burning aerosol.

Subject of Research: Modeling benzene and phenol oxidation pathways that form secondary organic aerosol

Article Title: Molecular representation of benzene and phenol secondary organic aerosols

Article References: Le Bayon, A., Ojala, A., Iyer, S., Wang, Z., Lannuque, V., Couvidat, F., Ciuraru, R., & Sartelet, K. (2026). Molecular representation of benzene and phenol secondary organic aerosols. Atmospheric Chemistry and Physics, 26(20), 14749-14768. https://doi.org/10.5194/acp-26-14749-2026

Image Credits: AI Generated

DOI: 10.5194/acp-26-14749-2026

Keywords: benzene oxidation, phenol oxidation, secondary organic aerosol, atmospheric chemistry, quantum chemistry, autoxidation, peroxy radicals, air-quality modeling, aerosol composition, GEMINAL diols, Molecular, representation

The bottleneck the team attacked is one that anyone who has pushed a weather or air quality model to higher resolution knows well. Doubling the horizontal resolution of an air quality model quadruples the number of grid cells, and numerical stability then forces the model to take proportionally shorter time steps. The combined effect is punishing: previous research cited in the study estimates that increasing horizontal resolution eighteenfold inflates the computational load three hundred times. Operational forecasting systems today mostly run at horizontal resolutions of several to tens of kilometers, which is too coarse for the urban-scale, street-by-street predictions that city managers increasingly demand. Faster hardware and smarter code are therefore not luxuries but prerequisites for the next generation of forecasts.

Graphics processing units offer a way out, because their thousands of parallel processors excel at exactly the kind of repetitive grid arithmetic that atmospheric models perform. The scale of modern GPU power is striking: the El Capitan supercomputer, which topped the TOP500 list in late 2024, delivers 1.742 exaflops of double-precision performance largely through AMD Instinct MI300A accelerators, each of which alone outperforms the entire Earth-Simulator, the world’s fastest machine in 2003. Yet most atmospheric models, including EPICC-Model, were written in Fortran for CPUs, with loop structures and data flows that translate poorly to GPUs. Simply copying code across rarely works; the algorithms must be rethought for the hardware.

The EPICC-Model team had already seen this dynamic play out within their own code. In earlier work, Cao and colleagues ported the model’s gas-phase chemistry module, historically the dominant cost at more than 40 percent of simulation time, onto China’s domestically developed GPU-like accelerators using the Heterogeneous-compute Interface for Portability, or HIP, a programming framework analogous to NVIDIA’s CUDA. That port cut the chemistry module’s share of runtime from 45.7 percent to 29.9 percent, but it exposed a new culprit: the horizontal advection module, which moves pollutants across the model grid with the wind, jumped from 13.3 percent to 25.0 percent of total runtime and became the model’s leading bottleneck. The new study targets exactly that module.

The heart of the advection scheme is the piecewise parabolic method, or PPM, a high-order Godunov-type algorithm from 1984 that approximates transported quantities with parabolic functions and is prized for its precision and robustness. It is the same scheme used in the latest versions of the widely used CMAQ and CAMx models. The researchers first rewrote the Fortran HADVPPM subroutine in standard C, then extended it with HIP calls so it could run on the GPU-like accelerator, which has 3840 computing units and 16 gigabytes of device memory, at the Earth System Numerical Simulation Facility (EarthLab) in Beijing. The resulting package, GPU-HADVPPM4HIP V1.0, follows the familiar heterogeneous computing sequence: allocate device memory, copy data from CPU to accelerator, launch parallel kernels, and copy results back.

Precision was checked at every step. Comparing the original Fortran, the rewritten C, and the HIP versions on identical inputs, the team found mean relative errors on the order of 10 to the minus 7 percent under default compilation, rising only slightly to around 10 to the minus 6 percent with aggressive O3 optimization. Interestingly, most of the small error crept in during the Fortran-to-C rewrite rather than the C-to-HIP step, which makes sense because HIP is essentially an extension of standard C. In a full 28-hour coupled simulation, the GPU version reproduced the Fortran model’s fields almost exactly, with differences in ozone of about 10 to the minus 2 ppbV and in PM2.5 of about 10 to the minus 3 micrograms per cubic meter, and normalized mean biases near zero.

The performance story unfolded in stages. A naive coupling, dubbed HIP-Ori, was actually a disaster: because the model called the advection solver roughly 9.8 million times per time step, each call triggering its own CPU-to-GPU data transfers, a one-hour integration ballooned from 1015 seconds in the original Fortran code to 20,400 seconds. The first fix, HIP-Opt1, restructured the loop hierarchy and expanded one-dimensional arrays into four- and five-dimensional blocks so that entire pollutant categories could be processed in single GPU calls, cutting transfers from about 9.8 million to just 10 per time step. That single change yielded a 17-fold improvement in model-level efficiency. A second optimization, HIP-Opt2, assigned each horizontal grid cell to its own GPU thread through coordinated thread-block indexing, after decoupling iterative calculations so that no step depended on the previous one, adding another 1.5-fold gain.

The team also implemented a hybrid MPI-plus-HIP scheme across EarthLab’s heterogeneous cluster, mapping each CPU process to a dedicated GPU accelerator so that the model scales across many nodes. In the final configuration, a 28-hour simulation that took the unoptimized GPU version 149.7 hours of wall-clock time finished in just 6.0 hours, roughly a 25-fold cumulative improvement over the naive port. At the module level, the GPU advection solver including data transfer overhead ran 39.3 percent faster than the Fortran original; with transfers excluded, the kernel itself was 20.5 times faster. Offline, on an array of 100 million double-precision values, the HIP solver compiled with O3 optimization reached that headline 556.5-fold speedup over the unoptimized Fortran baseline, and still managed 128-fold when both were compiled with the same aggressive flags.

Perhaps the most instructive finding is where the remaining time goes. Profiling showed that for large inputs, memory copying between CPU and GPU consumes roughly half of total execution time, a share that keeps growing with data size. This confirms that data movement, not arithmetic, is the defining constraint of heterogeneous computing, and it explains why the coupled model gains far less than the raw kernel speedup. The authors point to remedies including pinned-memory chunked transfers, asynchronous communication, unified memory architectures, and mixed-precision computation, as well as newer PCIe interconnects whose per-lane data rates have climbed from 16 gigatransfers per second in PCIe 4.0, used at EarthLab, to a projected 128 in PCIe 7.0.

The work also carries a broader message about scientific software in the GPU era. The EPICC-Model with its new advection solver reproduced observed PM2.5 and ozone variations reasonably well across eight Chinese cities during July 2021, with mean correlation coefficients of 0.70 for PM2.5 and 0.76 for ozone, demonstrating that the accelerated code remains scientifically trustworthy on domestic hardware. As the authors note, achieving kilometer-scale pollution simulations, of the kind that recently won a Gordon Bell Prize for climate modeling using NVIDIA GH200 GPUs, will require adapting entire air quality models to heterogeneous supercomputers, where nine of the world’s ten fastest machines now reside. Porting a single module is only the beginning; the real challenge, and opportunity, lies in redesigning the whole computational workflow around the GPU’s massively parallel architecture.

Subject of Research: GPU acceleration of the advection module in the EPICC-Model air quality modeling system

Article Title: Enhancing the advection module performance in the EPICC-Model V1.6.0 via GPU-HADVPPM4HIP V1.0 coupling and GPU-optimized strategies

Article References: Cao, K., Wu, Q., Tang, X., Li, J., Chen, X., Chen, H., Wang, W., Wu, H., Kong, L., Li, J., Zhu, J., & Wang, Z. (2026). Enhancing the advection module performance in the EPICC-Model V1.6.0 via GPU-HADVPPM4HIP V1.0 coupling and GPU-optimized strategies. Geoscientific Model Development, 19(19), 9807-9829. https://doi.org/10.5194/gmd-19-9807-2026

Image Credits: AI Generated

DOI: 10.5194/gmd-19-9807-2026

Keywords: GPU computing, air quality modeling, EPICC-Model, advection, piecewise parabolic method, HIP, high-performance computing, heterogeneous computing, EPICC-Model V1.6.0, GPU-HADVPPM4HIP, EarthLab, numerical weather prediction

The EPICC-Model, developed with support from the National Natural Science Foundation of China and the Earth System Numerical Simulation Facility, was officially released in November 2024. It is a three-dimensional tropospheric chemical transport model with a modular, plug-in architecture that simulates emissions, transport, gas-phase and heterogeneous reactions, aerosol thermodynamics, and dry and wet deposition. Several of its process routines and numerical algorithms were inherited or adapted from CAMx and CMAQ, whose foundational contributions the developers explicitly acknowledge, but the EPICC team independently integrated specialized chemistry treatments, including sulfate formation, heterogeneous nitrous acid, or HONO, production, and the hydrolysis of dinitrogen pentoxide. The new study, published in Geoscientific Model Development, represents the first comprehensive benchmarking of the model against established international systems under controlled conditions.

The experimental design is what gives the comparison its power. All three models were driven by identical meteorological fields generated with the Weather Research and Forecasting model and a harmonized multi-source emission inventory built around Tsinghua University’s MEIC inventory for China, supplemented by PKU-NH3 ammonia emissions, a biomass burning inventory, and EDGAR data for regions outside China. Biogenic emissions came from the MEGAN model run online. The simulation covered all of 2021 on a nested grid with 15-kilometer resolution over China, and the results were evaluated against hourly PM2.5 and ozone observations from 1,644 national monitoring stations across 337 cities. The WRF meteorology itself passed stringent validation, with temperature correlations above 0.7 in most regions and humidity errors within acceptable bounds, ensuring that differences among the air quality models reflected their internal chemistry and physics rather than divergent weather inputs.

For PM2.5, the verdict was nuanced. CAMx delivered the best overall statistics, with 94 percent of predictions falling within a factor of two of observations and a normalized mean bias of just minus 2.86 percent, though it consistently missed severe pollution episodes. CMAQ achieved the highest correlation, 0.85, but systematically overestimated concentrations, a bias the authors traced largely to inflated organic carbon levels linked to its Potential Combustion Secondary Organic Aerosol parameterization. The EPICC-Model landed in between, showing a pronounced seasonal response: it tracked CAMx closely during the low-concentration spring and summer months, then shifted toward CMAQ-like behavior in autumn and winter, capturing pollution accumulation more effectively. Spatially, the EPICC-Model achieved the highest annual index of agreement, 0.80, compared with 0.78 for CAMx and 0.77 for CMAQ, and it performed especially well in winter and over heavily polluted North China, where its index of agreement reached 0.81.

Ozone told a different story, and it is here that the new model made its most striking impression. All three models reproduced the seasonal cycle of the daily maximum 8-hour average ozone concentration, but during summer peaks, when photochemistry runs at full throttle, both CAMx and CMAQ systematically underestimated the highest values by as much as 20 micrograms per cubic meter. The EPICC-Model captured these peaks with markedly smaller bias and achieved the lowest annual normalized mean bias of the three. The authors attribute this advantage, at least in part, to the model’s inclusion of heterogeneous HONO formation and particulate nitrate photolysis, chemical pathways that supply additional hydroxyl radicals and accelerate ozone production over inland China. Recent sensitivity studies with the EPICC-Model support this interpretation, showing that these radical sources measurably improve simulated ozone formation across the country.

The model’s advantage extended into operational territory. When the researchers scored categorical forecasts of pollution events, the EPICC-Model posted the highest hit rate, 45.6 percent, for ozone pollution episodes exceeding 160 micrograms per cubic meter, ahead of CAMx at 43.0 percent and CMAQ at 38.4 percent, while maintaining a low false alarm rate of 3.8 percent. For general PM2.5 pollution above 75 micrograms per cubic meter, CMAQ led with a 64.7 percent hit rate but paid for it with the highest false alarm rate, 9.6 percent, a signature of its chronic overprediction. CAMx was the most conservative forecaster, with the fewest false alarms but also the most misses. The EPICC-Model struck a balance between detection and false alarms, and it showed competitive accuracy in the light and moderate pollution categories that matter most for day-to-day public health advisories.

The study also probed how well the models handle the hardest test in the business: persistent, multi-day pollution episodes that spread across entire regions. In a January 2021 PM2.5 event that formed a dual-core pattern over the North China Plain and the Sichuan Basin before sweeping southeast toward the Yangtze, all three models captured the arc from buildup to cold-front clearance, but the EPICC-Model best reproduced the spatial gradients and temporal evolution, particularly over the North China Plain and the middle Yangtze corridor. CAMx underestimated the intensity and missed the strong core over Henan at the peak, while CMAQ overestimated throughout. A June ozone episode that stretched from the North China Plain to Guangzhou revealed a similar pattern, with the EPICC-Model capturing the broad regional extent of elevated ozone, though it smoothed the enhancement and locally overestimated concentrations in central and southeastern China during the peak stage.

Beneath the headline scores, the comparison exposed chemical fault lines that run through all three models. Observations show nitrate now dominates PM2.5 mass in most Chinese cities, accounting for roughly 31 to 42 percent, and all models reproduced this nitrate-dominated regime. Yet every model systematically underestimated sulfate, likely because of insufficient sulfur dioxide oxidation rates or gaps in aqueous-phase chemistry, a shortfall that cascades into reduced ammonium and, paradoxically, can push nitrate even higher as leftover ammonia partitions with nitric acid. Organic carbon remained the most uncertain component, with normalized bias stretching from minus 50 to plus 150 percent in some cities, reflecting deep uncertainties in secondary organic aerosol formation, precursor emissions, and gas-particle partitioning. Black carbon, the least abundant component, was simulated most consistently, though slight overestimations in Beijing and Zhengzhou hinted at emission allocation issues.

The authors are candid about the shared weaknesses. All three models failed to reproduce heavy pollution episodes well, with hit rates for severe PM2.5 events above 150 micrograms per cubic meter ranging from a dismal 5.4 to 26.8 percent, even as false alarm rates stayed below 2 percent. All underestimated PM2.5 in the arid northwest because none of the operational configurations included wind-blown dust, and all overestimated autumn and winter concentrations in North China and the Sichuan Basin, a bias the study links to boundary-layer schemes that under-mix the nocturnal atmosphere under stable conditions. Emission inventory uncertainties compounded the problem, with anthropogenic emissions likely overestimated in the Sichuan Basin and agricultural ammonia underestimated in Henan, distorting nitrate formation in both regions. The comparison also noted that boundary conditions were not harmonized across the models, meaning some differences reflect inflow assumptions rather than internal chemistry.

What emerges from this benchmark is a portrait of a young model that has already earned a seat at the table. The EPICC-Model matches or beats its established rivals where it matters most for Chinese policy: summer ozone peaks, winter haze patterns over the North China Plain, and the life cycle of persistent compound pollution events, advantages that appear tied to its distinctive heterogeneous chemistry. At the same time, the study charts a clear roadmap for improvement, calling for refined secondary aerosol chemistry, better treatment of vertical turbulent mixing under stable boundary layers, incorporation of dust processes, and higher-resolution emission inventories with improved temporal variability. Because the framework, code, and data are openly available, the comparison sets a template for how new atmospheric models anywhere can be rigorously vetted, and it signals that the tools for managing China’s air, and by extension the air of other rapidly developing regions, are becoming both more diverse and more capable.

Subject of Research: Comparative evaluation of the EPICC-Model chemical transport model against CAMx and CMAQ for simulating PM2.5 and ozone pollution across China

Article Title: Evaluating the EPICC-Model for regional air quality simulation: a comparative study with CAMx and CMAQ

Article References: Lou, M., Wu, Q., Wang, W., Chen, H., Cao, K., Fan, X., Liang, D., Yu, F., Zhang, J., Wang, W., & Wang, Z. (2026). Evaluating the EPICC-Model for regional air quality simulation: a comparative study with CAMx and CMAQ. Geoscientific Model Development, 19(19), 9651-9676. https://doi.org/10.5194/gmd-19-9651-2026

Image Credits: AI Generated

DOI: 10.5194/gmd-19-9651-2026

Keywords: EPICC-Model, air quality modeling, chemical transport models, PM2.5, ozone pollution, CMAQ, CAMx, China haze, secondary aerosols, HONO chemistry, model intercomparison, WRF meteorology

The chemistry at the heart of the study begins after dark. Dinitrogen pentoxide, or N2O5, is a key player in nocturnal tropospheric chemistry, forming when nitrogen oxides react with the nitrate radical under dark conditions. When N2O5 collides with chloride-laden surfaces, whether airborne aerosol particles, saline snowpack, vegetation, buildings, or soil, it can undergo heterogeneous hydrolysis to produce nitryl chloride, ClNO2. This molecule acts as a reservoir for reactive chlorine: after sunrise, sunlight photolyzes ClNO2, releasing highly reactive chlorine atoms. Those chlorine radicals are ferocious oxidants, attacking alkanes at rates roughly two orders of magnitude faster than the hydroxyl radical, the atmosphere’s usual cleaning agent. The result is a cascade of reactions that can enhance the atmospheric oxidizing capacity and drive the formation of secondary pollutants, including fine particles and ground-level ozone.

What has remained murky is the role of snow itself. Field studies in high-latitude regions have previously confirmed that ice and snow surfaces catalyze the heterogeneous hydrolysis of N2O5 and that saline snowpack can emit an upward net flux of ClNO2. One-dimensional models have suggested that urban winter snowpack could contribute up to 60 percent of near-surface ClNO2. But such models struggle to capture the full complexity of air-snow exchange. The new work takes a decisive step forward by embedding surface chemistry into a fully three-dimensional air quality model, the WRF-CAMx system, allowing bidirectional fluxes between the atmosphere and the ground to be represented across an entire region rather than a single vertical column.

The study region could hardly be better suited to the question. Northeast China endures months of continuous snow cover, and its heating season stretches for nearly half the year. Large-scale coal burning during that period releases substantial quantities of particulate chloride and hydrogen chloride into the atmosphere. Some of these chlorides adsorb onto particle surfaces or deposit onto the snowpack, where they become feedstock for N2O5 reactions. To anchor the modeling, the team conducted field observations from 23 February to 3 March 2024 at a station on experimental farmland in the northern suburbs of Changchun, using an iodide-adduct chemical ionization mass spectrometer to measure N2O5 and ClNO2 amid recurrent snowfall that maintained continuous snow cover.

The observations revealed textbook diurnal cycles: N2O5 concentrations averaged 58.73 pptv and ranged as high as 424.15 pptv, while ClNO2 averaged 102.64 pptv with a peak of 902.45 pptv on 27 February. Both species accumulated at night and collapsed after dawn as photolysis destroyed ClNO2 and released chlorine atoms. Crucially, the data showed that ClNO2 formation was limited by the availability of N2O5 rather than by chloride supply. During one episode, when N2O5 mixing ratios stayed below roughly 120 pptv, no ClNO2 peaks appeared despite abundant chloride, evidence that the region’s chloride-containing aerosols provide a non-limiting reservoir of reactive chlorine throughout the winter.

On the modeling side, the researchers confronted a long-standing discrepancy between laboratory and field estimates of the N2O5 uptake coefficient, the probability that a colliding N2O5 molecule is irreversibly taken up by a surface. The widely used Bertram and Thornton parameterization, derived from chamber experiments, has consistently overestimated uptake compared with ambient measurements. The team therefore compared it against the YU20 scheme, which was fitted to field observations from multiple Chinese sites. Under snow-covered conditions, YU20 clearly outperformed BT09: for N2O5, the mean bias dropped from 114.62 to 69.66 pptv and the index of agreement rose from 0.49 to 0.61, while for ClNO2 the mean bias improved from −29.28 to −16.08 pptv. The researchers attribute the difference to the altered humidity and elevated anthropogenic chlorine emissions that reshape aerosol composition under snowy winter conditions, and they applied YU20 to snow-covered grid cells and BT09 elsewhere.

The most striking gains came from adding ground-surface chemistry. The team modified CAMx’s surface chemistry module to dynamically calculate surface-to-volume ratios from three-dimensional surfaces within each grid cell, drawing on field surveys and satellite retrievals of building areas across Northeast China, and to include N2O5 reactions on snow, soil, vegetation, and buildings. Before the revision, the model badly underestimated ClNO2, with a mean bias of −105.78 pptv and an index of agreement of just 0.39. After incorporating anthropogenic chlorine emissions and surface reactions, the mean bias shrank to 2.66 pptv and the index of agreement climbed to 0.86, while the root mean square error fell from 170.26 to 88.28 pptv. Simulated ClNO2 averages rose from 7.73 to 110.30 pptv, closely matching observations. Improvements rippled through secondary pollutants as well: for PM2.5 the mean bias improved from −10.56 to −1.39 micrograms per cubic meter, and for ozone from 19.42 to 13.91.

Quantifying the individual contributions, the researchers found that anthropogenic chlorine emissions alone raised ClNO2 by up to 151.67 pptv, with contributions exceeding 50 percent across most of the region, a larger share than the 30 to 50 percent reported in earlier snow-free modeling studies. The team links this amplification to snow’s high albedo, which boosts the photochemistry involved in ClNO2 generation. Ground-surface chemistry added a further 0 to 38.96 pptv, with contribution ratios reaching 48.20 percent in Heilongjiang, Jilin, and northeastern Liaoning. The day-night contrast was dramatic: nighttime surface-chemistry contributions peaked at 72.14 pptv, roughly ten times the daytime maximum, with ratios near 92 percent in parts of Heilongjiang and Jilin. Low temperatures favor the equilibrium shift of the N2O5–NO3 reaction toward N2O5, and long winter nights allow dark accumulation, while snow’s large reactive interface and moisture favor gas deposition and heterogeneous reaction.

These chemical changes translated into measurable air quality impacts. Chlorine chemistry increased hourly PM2.5 concentrations by up to 3.65 micrograms per cubic meter, a relative contribution of 15.34 percent, concentrated around cities such as Harbin and Changchun, while promoting ammonium and sulfate particle formation and suppressing particulate nitrate by as much as 23.77 percent. Maximum daily 8-hour average ozone rose by up to 3.41 ppbv, or 5.68 percent. The simulations also showed chlorine chemistry reshaping the atmospheric oxidizing capacity, promoting OH generation with relative contributions up to 7.59 percent, consistent with earlier estimates of chlorine’s role in regional oxidation.

The authors caution that their surface chemistry module remains simplified, focusing on re-emission from predefined reactions, and that the behavior of snowmelt, where shifting liquid water content in snow and ice pores alters reaction rate constants, deserves particular attention. They propose establishing quantitative relationships among snow depth, total ion concentration, and liquid brine fraction through field sampling and laboratory experiments, which would allow uptake coefficients and ClNO2 yields to be pinned down for different snow depths. Even so, the study delivers a clear message: frozen landscapes are chemically active players in winter air pollution, and accurately representing the flux exchange between the atmosphere and the cryosphere is essential for forecasting, and ultimately mitigating, the smog that blankets cold regions each winter.

Subject of Research: Heterogeneous chlorine chemistry on snowpack surfaces and its role in secondary atmospheric pollution in Northeast China

Article Title: Numerical modeling on the mechanisms of chlorine chemistry in snowpack and their impact on secondary atmospheric pollution

Article References: Xie, S., Zhang, X., Xiu, A., Qi, H., Tong, S., Chen, Q., Gao, C., Zhao, H., Zhang, S., Yabo, S. D., Liu, Y., Li, S., & Zhang, M. (2026). Numerical modeling on the mechanisms of chlorine chemistry in snowpack and their impact on secondary atmospheric pollution. The Cryosphere, 20(9), 5435-5451. https://doi.org/10.5194/tc-20-5435-2026

Image Credits: AI Generated

DOI: 10.5194/tc-20-5435-2026

Keywords: snowpack chemistry, nitryl chloride, dinitrogen pentoxide, chlorine radicals, WRF-CAMx, air quality modeling, PM2.5, ozone, Northeast China, heterogeneous reactions, atmospheric oxidizing capacity, The Cryosphere

The study, led by Jesse Bash of the US Environmental Protection Agency, now at the Norwegian Meteorological Institute, together with colleagues from the EPA, Environment and Climate Change Canada, and partner institutions, scrutinized the Surface Tiled Aerosol and Gaseous Exchange option, known as STAGE, in version 5.3.2 of the Community Multiscale Air Quality model, or CMAQ. STAGE is a unified framework that estimates both dry deposition, the uptake of gases by surfaces, and bidirectional exchange, the two-way traffic of gases between the atmosphere and soils or vegetation. What makes STAGE distinctive is that it applies the same resistance model to every modeled gas, whether that gas only deposits, like nitric acid, or can flow in both directions, like ammonia. That consistency matters, because in earlier versions of CMAQ, ammonia bidirectional exchange and the deposition of all other species were handled by two separate, parallel resistance schemes with different structural assumptions.

The mathematical heart of STAGE is a resistance analog borrowed from electrical circuit theory, following the widely used parameterizations of Nemitz and colleagues from 2001 and Massad and colleagues from 2010. The flux of a trace gas is driven by the difference between the ambient concentration and a compensation point concentration at the surface, divided by the aerodynamic resistance. For ammonia, the compensation point is built from a two-layer canopy and soil model: gases diffuse through leaf boundary layers, stomata, cuticles, and in-canopy air, with each pathway assigned its own resistance. The model even solves for the leaf-level compensation point using a system of equations analogous to Kirchhoff’s current law, balancing exchange between the atmosphere, stomata, cuticular surfaces, and the ground. When the compensation points are set to zero, the framework collapses gracefully into a conventional dry deposition velocity, which is precisely what allows one model to serve both unidirectional and bidirectional species.

Several technical refinements were introduced to make this framework work at regional scale. Deposition to wet leaf surfaces, previously treated as instantaneous diffusion into canopy water, now uses a mass transfer formulation adapted from CMAQ’s aqueous chemistry module, accounting for gaseous diffusivity and bulk accommodation into roughly spherical droplets sitting on hydrophobic leaves. The depth of surface moisture is derived from relative humidity using Brunauer-Emmett-Teller adsorption theory calibrated against leaf wetness observations, and droplet radius follows from that water depth and the leaf area index. In-canopy aerodynamic resistance is computed by integrating eddy diffusivity through the canopy using a momentum attenuation approach. Soil processes were also updated: the diffusive length for soil gas exchange was increased to two centimeters to match the measurements underlying the model, and soil ammonium availability is now estimated with a nonlinear sorption curve rather than a fixed solution fraction.

To test whether all this machinery reflects reality, the team turned the regional model into a field-scale box model written in the R statistical language, using identical resistance parameterizations, and ran it against micrometeorological flux measurements from two North Carolina sites. The first was a managed, unfertilized 15-hectare grass field at Duke Forest, where fluxes of ammonia, nitric acid, and sulfur dioxide were measured with a MARGA instrument during the second half of 2012. The second was a 200-hectare fertilized corn field near Lillington, where ammonia gradients were measured with a continuous-flow wet denuder system in 2007. When the box model was fed observed soil and vegetation ammonia emission potentials, the compensation points that quantify how much ammonia the soil and plant tissues are ready to release, the modeled fluxes of all three gases agreed well with observations, with mean biases within or near the reported measurement uncertainty.

The trouble emerged when the model was run with the default tabular values built into CMAQ. At the Duke Forest grassland, the defaults for soil and vegetation emission potentials, a soil gamma of 20 and an apoplast gamma of 247, produced a simulated mean net deposition of minus 1.3 nanograms per square meter per hour, while the towers actually recorded a mean evasive flux of 8.4 nanograms per square meter per hour in the opposite direction. In plain terms, the model said the grassland was absorbing ammonia when it was actually emitting it. Sensitivity experiments showed the modeled flux was most sensitive to the vegetation emission potential, and that perturbing resistances alone could not reverse the sign of the error. The measured compensation points at the site simply exceeded the tabular assumptions, which were drawn from parameterizations based on annual nitrogen deposition fields for non-agricultural land.

The corn field told a more encouraging but nuanced story. Using median measured soil ammonium, soil pH, and an apoplast emission potential of 153.5 reported in earlier work, the model captured the observed fluxes with a normalized mean bias of minus 23.4 percent, rising to 40.5 percent with the default parameterizations. The site also displayed a persistent mid-morning emission peak that the model could not fully explain. Dew on the leaves carried a high emission potential, but there was not enough ammonium in the dew, and the dew dried too early, to account for emissions that continued between eight and eleven in the morning on a dry canopy. The authors speculate that wetting and drying of ammonium-rich exposed soil between the rows, a process the soil moisture probes missed, may drive the morning pulse, contrasting with earlier dew-evaporation explanations at a Colorado grassland.

The decisive test came at continental scale. The team compared an annual 2016 CMAQ simulation against ammonia retrievals from the Cross-Track Infrared Sounder, a satellite instrument capable of detecting ammonia down to roughly half a part per billion under typical conditions. The model captured the broad magnitude and spatial patterns of satellite-observed ammonia, but the biases were revealing. CMAQ overestimated concentrations by 0.5 to 0.9 parts per billion on average in intensely agricultural regions such as California, the Upper Midwest, and eastern North Carolina, while underestimating observations across the Great Plains and other vegetated areas, with a broad shortfall of roughly one to two parts per billion over non-agricultural land. That pattern mirrors the grassland box model result almost exactly, pointing to a common cause: default emission potentials for natural vegetation and soils that sit at the low end of what field measurements show.

The implications extend beyond model diagnostics. The team estimates that of the 10.3 megatonnes of nitrogen applied as fertilizer to row crops in the 2016 simulation, about 0.5 megatonnes, an emission factor of 4.8 percent, was re-emitted as ammonia, a figure consistent with published ranges. Deposition peaked near 30 kilograms of nitrogen per hectare downwind of agricultural sources in eastern North Carolina, within measured values but higher than some earlier net deposition estimates for the region. The authors conclude that raising the grassland vegetation emission potential by about 30 percent and the soil surface emission potential roughly fivefold would better match both the Duke Forest observations and the satellite data, changes that remain well within the natural variability reported across measurement studies. Because non-agricultural land covers roughly three quarters of the terrestrial surface in the model domain, the path forward is clear: more micrometeorological flux measurements and soil and vegetation chemistry across different land uses, soil types, and growth stages. The multi-resolution evaluation strategy, spanning flux towers, monitoring networks, and satellites, offers the modeling community a template for building the next generation of air-surface exchange models on firmer empirical ground.

Subject of Research: Evaluation of bidirectional ammonia, nitric acid, and sulfur dioxide air-surface exchange in the CMAQ model's STAGE deposition option against field, network, and satellite observations

Article Title: Evaluation of HNO3, SO2, and NH3 in the Surface Tiled Aerosol and Gaseous Exchange (STAGE) option in the Community Multiscale Air Quality Model version 5.3.2 against field-scale, in situ and satellite observations

Article References: Bash, J. O., Walker, J. T., Wu, Z., Rumsey, I. C., Murphy, B., Hogrefe, C., Fahey, K. M., Pye, H. O. T., Jones, M. R., Appel, K. W., Shephard, M. W., Alnsour, N. I., & Cady-Periera, K. E. (2026). Evaluation of HNO 3 , SO 2 , and NH 3 in the Surface Tiled Aerosol and Gaseous Exchange (STAGE) option in the Community Multiscale Air Quality Model version 5.3.2 against field-scale, in situ and satellite observations. Geoscientific Model Development, 19(19), 9377-9393. https://doi.org/10.5194/gmd-19-9377-2026

Image Credits: AI Generated

DOI: 10.5194/gmd-19-9377-2026

Keywords: ammonia, air quality modeling, CMAQ, STAGE, dry deposition, bidirectional exchange, satellite observations, CrIS, grassland flux, emission potential, nitrogen deposition, particulate matter

The team, led by Yuan Dai and corresponding authors Junfeng Wang of Nanjing University of Information Science and Technology and Xinlei Ge of Southeast University, exploited a natural experiment created by the COVID-19 pandemic. They compared emission control periods, when lockdowns sharply curtailed traffic and industry, with adjacent normal periods in both winter and summer. During these control periods, nitrogen oxide concentrations fell by 39 percent in winter and 34 percent in summer, while sulfur dioxide barely moved, staying between 3.5 and 3.9 parts per billion. That asymmetry pushed the gas-phase sulfur-to-nitrogen ratio up by 71 percent in winter and 60 percent in summer, while ozone rose about 15 percent because less of it was being destroyed by fresh nitrogen oxide emissions.

What makes the study unusual is its single-particle approach. Instead of averaging the composition of all particles together, as bulk measurements do, the researchers used a single-particle aerosol mass spectrometer to classify each particle by its dominant chemical signature. They sorted the population into black carbon particles, black carbon mixed with organic carbon, organic-rich particles without black carbon, and black-carbon-free inorganic particles, then subdivided these into 19 subclasses based on sulfate, nitrate, and cyanide-related fragments. Fresh combustion particles, marked by cyanide ions and small sizes of roughly 100 to 600 nanometers, could be distinguished from aged particles coated in secondary sulfate and nitrate.

By defining three complementary sulfur-to-nitrogen ratios, one for the gas phase, one for the particle phase, and one based on the relative numbers of sulfur-containing versus nitrogen-containing particles, the team could watch the same chemical signal move through different domains of the atmosphere. During emission control periods, the particle-phase ratio rose by roughly 57 percent on average across particle classes, and the mean vacuum aerodynamic diameter of particles grew by 27 percent. In other words, when the air became relatively sulfur-rich, particles not only changed their internal chemistry but also grew larger, likely through the accumulation of sulfate coatings.

The most striking finding concerns which particles respond most strongly. Particles containing both black carbon and organic carbon showed the greatest sulfate enrichment relative to black-carbon-free inorganic particles, in both seasons. In summer, the enhancement factor for black carbon particles climbed from 1 to 3 in the morning to 6 to 9 near solar noon, suggesting that photochemically active hours favor sulfate formation on soot surfaces, possibly through black-carbon-mediated release of nitrous acid and nitrogen dioxide and light-induced reactive oxygen species that accelerate the conversion of sulfur dioxide to sulfate. At night, when humidity exceeded 80 percent and sunlight vanished, organic-rich particles took over as the dominant site of sulfate enrichment, consistent with aqueous-phase oxidation by hydrogen peroxide and ozone and with the formation of organosulfates.

Mixed black carbon and organic carbon particles showed the largest enhancement of all, in both day and night and in both seasons. The authors propose a synergistic mechanism: the black carbon fraction provides reactive or catalytic surfaces, while the organic fraction helps sulfate accommodate and remain trapped within the particle matrix. Light-absorbing organic compounds such as brown carbon may even act as photosensitizers, speeding up sulfur dioxide conversion under ultraviolet light. In winter, when photochemistry is weak, organic-driven aqueous pathways dominate, yet the mixed particles still hold their advantage, possibly aided by the larger accessible surface area of soot’s tangled morphology and by coating softening that promotes sulfate retention.

Humidity emerged as the master regulator of this entire cross-phase coupling. The researchers divided their observations into four relative humidity regimes anchored to known phase transitions of ammonium nitrate and ammonium sulfate, the two dominant secondary inorganic salts in fine particulate matter. Below 55 percent relative humidity, particles are largely solid, gas-particle exchange is kinetically limited, and the three sulfur-to-nitrogen ratios decouple from one another. The particle population remains externally mixed, with substantial chemical differences between particle types. Crossing the deliquescence point of ammonium nitrate near 61 percent humidity triggers a semi-solid transition that activates particle surfaces and pushes the particle-phase ratio upward, especially for black-carbon-containing classes.

Above 85 percent relative humidity, the picture changes dramatically. Particles take up enough water to become predominantly aqueous, kinetic barriers to gas-to-particle partitioning collapse, and all three sulfur-to-nitrogen ratios converge to similarly low, stable values, with correlations between them reaching coefficients of 0.68 to 0.81. The differences between particle types nearly vanish as the aerosol evolves toward an internally mixed state, in which water-mediated multiphase processing homogenizes composition across the population. This humidity-dependent transition from chemical heterogeneity to uniformity is precisely the kind of behavior that models treating particles as a well-mixed average cannot capture.

To move beyond correlation, the team combined causal inference with interpretable machine learning. A causal network analysis, using the Grow-Shrink and Greedy Equivalence Search algorithms, identified directed links running from the gas-phase ratio to the particle-phase ratio and onward to the number-based ratio, with ozone and boundary layer height acting as intermediaries between the chemistry and the meteorological environment. A machine-learning model, an XGBoost framework optimized by a Sparrow Search Algorithm, achieved coefficients of determination between 0.71 and 0.91 across particle classes. Shapley value analysis confirmed that the gas-phase ratio was the dominant driver of particle-phase partitioning, and that the particle-phase ratio in turn led the number-based ratio, supporting the sequential propagation picture. Relative humidity exerted a consistent negative effect, while solar radiation and ozone contributed positively.

The implications reach beyond Chinese cities. As clean air policies worldwide tighten nitrogen oxide controls faster than sulfur dioxide controls, the gas-phase sulfur-to-nitrogen ratio may rise again, potentially alongside higher ozone levels and a more oxidizing atmosphere. The study shows that such shifts will not affect all particles equally: soot particles will accumulate sulfate coatings differently than organic or purely inorganic ones, with knock-on consequences for how particles take up water, scatter light, and act as cloud condensation nuclei. The authors argue that future air-quality models should incorporate particle-type heterogeneity and humidity-dependent multiphase processes, rather than relying on bulk averages, if they are to predict how secondary aerosols will evolve under the emission regimes of the coming decades.

Subject of Research: Cross-phase sulfur-nitrogen partitioning and aerosol mixing-state evolution observed at the single-particle level

Article Title: Cross-phase partitioning of sulfur-nitrogen ratios and aerosol mixing-state evolution based on single-particle observations

Article References: Dai, Y., Wang, J., Chen, M., Zhang, S., Li, H., Zhang, Y., Wu, Y., Wang, M., & Ge, X. (2026). Cross-phase partitioning of sulfur-nitrogen ratios and aerosol mixing-state evolution based on single-particle observations. Atmospheric Chemistry and Physics, 26(19), 14015-14029. https://doi.org/10.5194/acp-26-14015-2026

Image Credits: AI Generated

DOI: 10.5194/acp-26-14015-2026

Keywords: aerosols, single-particle mass spectrometry, sulfur-to-nitrogen ratio, secondary aerosol formation, black carbon, relative humidity, mixing state, nitrogen oxides, sulfur dioxide, COVID-19 lockdown, air quality modeling, multiphase chemistry

The research team focused on Tianjin, a sprawling port megacity on the Bohai Bay in northern China, during the period from 2017 to 2020. These years coincided with aggressive national emission reduction programs, making the city an ideal natural laboratory for separating the effects of pollution controls from those of weather. The scientists combined ground-based chemical observations with the WRF-NAQPMS modeling system, a coupled meteorology and air quality model widely used in China for forecasting and policy analysis. By running the model with fixed meteorology and varying emissions, and vice versa, they could disentangle how much of the observed change in each pollutant came from human activity and how much came from the atmosphere itself. This decomposition technique, often called a factor-separation or scenario approach, is one of the most rigorous tools available for attributing air quality trends, and the team applied it not just to annual averages but to distinct weather regimes defined by the sea-land breeze cycle.

The headline numbers from the study are striking in themselves. Emission reductions between 2017 and 2020 drove PM2.5 concentrations down by 7.1 micrograms per cubic meter, a decline driven primarily by reductions in carbonaceous components and sulfate, the chemical fingerprints of combustion and industrial sources. Yet over the same period, ozone rose by 24.5 micrograms per cubic meter according to the model, closely matching the net observed increase of 23.0 micrograms per cubic meter. Meteorology, averaged across the whole year, played a comparatively minor role, contributing a decline of 1.7 micrograms per cubic meter for PM2.5 and 1.4 micrograms per cubic meter for ozone. On paper, this looks like a familiar story: cleaner particles, dirtier air in a different form, with weather as a footnote. But the averages, the researchers found, were hiding the real drama.

When the team broke the data down by circulation regime, the modest annual meteorological contributions dissolved into wildly divergent day-by-day realities. On sea breeze days, when cool marine air pushes inland during the afternoon, meteorology was the only regime in which it actually increased ozone, adding 2.2 micrograms per cubic meter. More remarkably, the sea breeze offset 5.1 micrograms per cubic meter of the PM2.5 reductions that emission controls had achieved, effectively erasing a substantial fraction of the hard-won progress. During these periods, the researchers also observed concurrent increases in nitrate and elemental carbon, indicating that the marine air flow was not simply diluting pollution but actively reshaping its chemical composition, likely by transporting and transforming aerosol precursors along the coast. The sea breeze, in other words, is not a cleansing wind; it is a chemical conveyor.

The land breeze told the opposite story with even greater intensity. On land breeze days, when the nocturnal flow reverses and carries air from the city out toward the sea, interannual meteorological changes favored PM2.5 reduction with a contribution of 32.70 micrograms per cubic meter, a figure nearly five times larger than the annual mean decline. This meteorological assist was accompanied by substantial reductions in nitrate and elemental carbon, confirming that the ventilation effect was removing real pollution rather than merely redistributing it. Then there were the stagnant, stable days, the bane of every air quality manager. Under these conditions, meteorological factors were the primary driver of interannual PM2.5 increases, contributing a staggering 30.80 micrograms per cubic meter, and PM2.5 peaks exceeded 150 micrograms per cubic meter even with emission reductions in full force. On such days, the atmosphere itself overwhelmed the benefits of a decade of controls.

The diurnal asymmetry the study documents adds another layer of complexity. Daytime sea breezes dominated ozone production, because the marine air arriving over the sun-heated land encounters a shallow, turbulent boundary layer in which photochemistry runs at full throttle. Precursor gases trapped near the surface are efficiently mixed and cooked into ozone under intense solar radiation. At night, the land breeze stage favored pollutant accumulation and secondary formation, as weak winds and temperature inversions allowed emissions to pool near the ground while heterogeneous chemistry continued to manufacture secondary aerosols in the dark. This means that a single 24-hour cycle in a coastal city contains two chemically distinct pollution environments, each responding differently to the same emission inventory. A policy calibrated against daily averages will inevitably misjudge both.

Why does ozone rise even as its precursors fall? The answer lies in the nonlinear photochemistry of the troposphere. Ozone formation depends on the ratio of nitrogen oxides to volatile organic compounds, and in many Chinese cities the chemical regime is what atmospheric scientists call volatile organic compound limited or transitional. When nitrogen oxide emissions drop, ozone destruction by nitrogen oxides weakens faster than ozone production, so ozone concentrations climb even though the ingredients for smog are diminishing. This VOC-limited behavior has been documented across the North China Plain and is a central reason why the co-control of PM2.5 and ozone has become the defining challenge of Chinese air quality policy. The Tianjin study adds a crucial spatial dimension to this problem: the sea-land breeze modulates the local chemical regime on a daily cycle, meaning the effective sensitivity of ozone to emissions can flip between morning and afternoon.

The implications for policy are profound. The study’s central conclusion is that regime-specific meteorological contributions dwarf the annual averages used in most policy evaluations, and that sea-land breeze circulation therefore fundamentally modulates pollutant responses to emission controls. The authors argue for circulation-specific co-control strategies in coastal megacities. In practice, this could mean scheduling stricter temporary emission curbs on forecast sea breeze days, when ozone formation will be amplified and particulate reductions partially offset, while relaxing controls on land breeze days when natural ventilation does much of the work. It could also mean targeting volatile organic compound reductions more aggressively than nitrogen oxides in coastal zones, to push the photochemical regime toward one where further nitrogen oxide cuts actually lower ozone. Cities from the Pearl River Delta to the Yangtze River Delta, where sea breeze effects on coastal ozone are already documented, could apply the same framework.

There is also a warning embedded in the findings about the future. Related research has shown that ocean warming and urbanization are weakening sea-land breezes in coastal megacities, which means the natural ventilation and chemical transport patterns that the Tianjin study quantified are themselves changing. A control strategy optimized for today’s breeze climatology may misfire as the circulation shifts. Moreover, the study’s demonstration that stagnant conditions can push PM2.5 above 150 micrograms per cubic meter despite emission reductions underscores that meteorological variability remains a first-order driver of extreme pollution events, not a second-order correction. Climate change is expected to alter the frequency and persistence of such stagnant episodes across many densely populated coastal regions, adding yet another moving part to an already intricate system.

What makes this study resonate beyond atmospheric science is its reframing of a familiar frustration. Residents of coastal cities have long sensed that some breezy days feel fresh and others feel thick with haze, and now there is a quantitative framework explaining why. The daily dance between sea and land is not background noise in the air quality record; it is a key modulation, capable of amplifying or erasing the effects of policy at scales that rival the policies themselves. For the growing number of megacities clustered along the world’s coastlines, from Tianjin to Mumbai to Lagos, the message is clear: to manage air pollution in a coastal city, you must first learn to read the wind.

Subject of Research: How sea–land breeze circulation modulates PM2.5 and ozone responses to emission reductions in coastal megacities

Article Title: Key modulation of PM2.5–O3 responses to emission reductions by sea–land breeze in coastal megacities

Article References: Liu, K., Wu, L., Lu, M., Tang, X., Kong, L., Zhang, Y., & Wang, Z. (2026). Key modulation of PM2.5–O3 responses to emission reductions by sea–land breeze in coastal megacities. Air Quality, Atmosphere & Health, 19(10), Article 220. https://doi.org/10.1007/s11869-026-02038-w

Image Credits: AI Generated

DOI: 10.1007/s11869-026-02038-w

Keywords: PM2.5, ozone, sea–land breeze, emission reductions, coastal megacities, air quality modeling, WRF-NAQPMS, Tianjin, photochemistry, meteorology, co-pollution, co-control strategies