We utilize the FLEXible PARTicle dispersion model (FLEXPART) version 10.4 (Pisso et al., 2019), using meteorology from the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5). ERA5 is a fifth-generation ECMWF atmospheric meteorological reanalysis of the global atmosphere at hourly temporal resolution (Hersbach et al., 2020). Hourly and monthly data were available on a 0.25° longitude × 0.25° latitude grid with 137 vertical levels ranging from 1000 to 1 hPa. We utilized 3-D wind (u,v,w), temperature, pressure, and other surface quantities (Cui et al.,2025).

FLEXPART–ERA5 generated the source–receptor relationship (SRR) quantity used in this study, which can be applied to represent the air mass residence time at specific times and grid locations. Thus, this is often referred to as the “surface influence factor”, “residence time”, or “retroplume” (e.g., Cooper et al., 2010; Cui et al., 2025). In this study, we refer to the SRR as residence time. A high SRR indicates a strong connection between the upwind source region and the observation location. These high SRR values indicate long residence time over source regions, resulting in greater surface emission influence, while a low SRR represents a weak connection between source regions and observations due to short residence times. The model is used to support our 28-year O3 analysis from 1994 to 2021, providing a set of outputs at hourly resolution extending from the observational O3 receptors back 15 d, with 1° × 1° spatial resolution globally (Cui et al., 2025). To track the vertical range of the SRR of trajectories associated with the O3 observation over WNA, we classify the trajectory layers into five vertical levels: 0–300 m, 300 m–3 km, 3–8 km, 8–13 km, and 13–20 km. These five levels refer to the retroplume assessment levels associated with observed O3 values over WNA, regardless of the original O3 measurement altitude. While the raw data contain detailed receptor and retroplume information (Cui et al., 2025), only retroplume data are retained when processing daily and monthly data.

The five vertical layers from the surface to 20 km are designed to support the investigation of the different sources associated with different altitudes and to understand their source contributions. To differentiate between source contributions, we utilize specific vertical layers: the 0–300 m and 0.3–3 km ranges quantify the respective contributions of surface-level and planetary boundary layer (PBL) emissions. The 3–13 km column accounts for the influence of lightning-induced ozone formation, while the 8–13 km range isolates the impact of aircraft emissions on downwind WNA O3 levels. The output unit of the SRR from the FLEXPART–ERA5 backward model simulation is sm3kg-1, which represents the SRR weighted by the air mass volume. For more detailed model processing information, please see Cui et al. (2025).

Daily output with 15 d back-trajectories is aggregated into monthly data by normalizing the total SRR by the O3 receptor observations. Although our initial daily output spans the globe, we restrict our monthly analysis to the NH to reduce computational demands. We justify this by assuming that interhemispheric transport contributes relatively little to O3 levels over WNA. To represent different atmospheric layers, the 0–300 m and 300 m–3 km ranges were categorized as the lower troposphere, the 3–8 km range as the free troposphere, the 8–13 km range as the upper troposphere and lower stratosphere (UTLS), and the 13–20 km range as the stratosphere in the tropics and mid-latitude region.

We used a data fusion technique to combine multi-platform O3 observations across 900–300 hPa above WNA (25–55° N, 130–90° W, Fig. 1a–c), to establish our receptor list over WNA (Chang et al., 2023a; Cui et al., 2025). Tropospheric O3 observations over WNA during 1994–2021 used in this study include (1) Ozonesonde measurements provided by the Canadian Ozonesonde Network (Environment and Climate Change Canada, 2022), and from the NOAA Global Monitoring Laboratory (NOAA GML, 2022), (2) Lidar measurements from Table Mountain (NASA JPL 2022), and (3) Aircraft measurements from IAGOS (In-Service Aircraft for a Global Observing System; Boulanger et al., 2022) and NASA AJAX/SNAX (Iraci and Yates, 2021; Yates et al., 2023). The availability of the observation periods varies slightly among the measurement sites during our study period from 1994–2021 (please refer to Fig. 1 of Cui et al., 2025 for more details). These efforts aim to enhance accuracy and precision of O3 trend estimates, building on earlier findings that sparse ozonesonde sampling – typically once per week – is insufficient for capturing accurate monthly means (Logan, 1999) or detecting reliable trends (Prinn, 1988). These findings were further evaluated and validated by using the densely sampled IAGOS dataset above Frankfurt, Germany (Chang et al., 2020; Saunois et al., 2012). More recently, Chang et al. (2024) demonstrated that trend estimates can vary significantly with sampling strategies, showing notable biases under sparse conditions. Gaudel et al. (2024) likewise emphasized the difficulty of detecting O3 trends in the tropics due to limited in-situ observations.

We focus on four major source regions affecting the levels over WNA: SE Asia (0–25° N, 60–130° E), NE Asia (26–46° N, 75–127° E), Siberia (50–75° N, 70–160° E), and Tropical Pacific Ocean (hereafter Tropical Pacific; 5–35° N, 180° E–130° W) because, (1) these regions reflect differing emission distributions linked to growing industries and emission control strategies (Zhang et al., 2008; Wang et al., 2019), (2) these regions are well known for significantly contributing to O3 precursor sources and sinks (Thorp et al., 2021), and (3) hemispheric-scale O3 production and loss and its transport (Jacob et al., 1999; Cooper et al., 2010), and biomass burning emissions transport acting as O3 precursor species often affect WNA through long-range transport (Johnson et al., 2021) from various vertical ranges (Ryoo et al., 2017). Although our “SE Asia” domain (0–25° N, 60–130° E) spans South Asia (including India and Pakistan), Southeast Asia (Indochina and maritime states), and southern East Asia, this macro-scale boundary was intentionally chosen to capture broad, interconnected transport pathways. Strong transboundary mixing across these neighboring zones (e.g., Samphutthanont et al., 2026) makes separating individual sub-regional signatures difficult over multi-day trajectories. Therefore, this domain is designed to capture the overarching regional transport footprint and ensure solid statistical sampling, rather than pinpoint localized source areas. In this analysis, we also selected the Tropical Pacific as a potential source region due to high SRR and close proximity to WNA. For examining anthropogenic shipping emissions, we categorize the domain into three oceanic regions – Tropical Indian (0–30° N, 60–150° E), Tropical Pacific (5–35° N, 180° E–130° W), and mid-latitude western Pacific (30–55° N, 127–152° E). More details are shown in Sect. 3.3.

GMI O₃ chemical production and loss data. we utilized the GMI data for O3 (Fig. S1 in the Supplement), NO2 (Fig. S2), and chemical O3 production and loss to examine net O3 production (Rotman et al., 2001; Ziemke et al., 2019). The GMI simulations employed the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2; Gelaro et al., 2017) reanalysis meteorology and a combined stratospheric-tropospheric chemical mechanism. GMI simulations have been widely used in various tropospheric studies to interpret observations of O3 and NOx, analyze observed trends (Ziemke et al., 2019), and investigate processes influencing O3 (Kerr et al., 2021; Strode et al., 2015, 2016). We focused on the period from 1994 to 2019. The spatial resolution is 0.625° longitude by 0.5° latitude and monthly temporal resolution.

Data from the GMI model and TCR-2 chemical reanalysis were applied to assess trends in O3 (Fig. S1) and its precursor species such as NOx (Figs. 1d and S2).

The number of receptors varies across years due to sampling differences in observational data availability. To account for this variability, we normalized the monthly SRRs by dividing the total SRRs from the receptors by the total receptor count for each month across different altitudes. Monthly O3 concentration percentiles from 1994 to 2021 were then defined relative to the monthly percentiles determined during the 2004–2014 baseline period. This period was selected as a representative baseline for O3 values at receptors over WNA because O3 sampling during this time was relatively evenly distributed.

We categorized O3 values based on the following percentiles: O3 < 5th percentile, 5th ≤ O3 < 33rd percentile, 33rd ≤ O3 < 50th percentile, 50th ≤ O3 < 66th percentile, 66th ≤ O3 < 95th percentile, and O3 ≥ 95th percentile. We defined low O3 using the 33rd percentile threshold, as SRR patterns were largely similar to those below the 5th percentile. To avoid the air masses with the strongest impact from stratospheric intrusion, we also excluded the 95–100th percentiles in our analysis.

The total number of monthly non-zero SRRs across the full range of O3 values from 1994 to 2021 was N = 539 808 out of 553 608 total gridded O3 observation (Cui et al., 2025). Of this total, the number of SRRs associated with O3 values below the 5th percentile and the 33rd percentile is N = 26 858 and 173 538, respectively, while those above the 66th percentile total N = 187 597, across all vertical layers. Those for high O3 (66th–95th percentile), excluding the highest O3 levels that are typically associated with recent stratospheric intrusions, is N = 160 299.

To analyze trends in airmass transport associated with particular O3 levels above WNA, as represented by SRR, we applied a linear regression model. For each month, the monthly SRR values corresponding to the O3 percentile ranges, as described in Sect. 2.1, were input into the linear regression model, formulated as follows: 1SRR(x,y,z)=M(x,y,z)+τ(x,y,z)×t+ϵ where x is longitude, y is latitude, z is the five vertical layers, M(x,y,z) is a monthly mean cycle, τ(x,y,z) is the trend coefficient, which is shown in the results, t represents the total 336 months (12 months × 28 years), and ϵ represents the residual term. The p-value of the trend estimate was used to assess the trend reliably. We adopted the TOAR statistical guidelines where a trend with p-value ≤ 0.05 is considered to be high certainty (Chang et al., 2023b; Wasserstein et al., 2019).

Monthly average SRRs (M) were first calculated for the five vertical layers at each longitude and latitude grid point from 1994 to 2021. Trends were then determined using a deseasonalized dataset, by subtracting the 28-year mean for each month (M(x,y,z)) from the corresponding monthly SRR (e.g., for January 2010, the anomaly was calculated by subtracting the average of each January from 1994 to 2021 from the January 2010 value). Similarly, the trend of net GMI O3 production is also computed following Eq. (1). Net O3 production was estimated by subtracting all O3 loss terms from all O3 production terms, based on the GMI dataset.

The percentage change in chemical net O3production (ΔO3_netp) is calculated in Eq. (2) based on the difference in the averages of the two periods (P1: 1994–2006 and P2: 2007–2019) normalized by their combined average during the P1 period: 2ΔO3_netp=O3_netp(P2:2007–2019)-O3_netp(P1:1994–2006)abs(O3_netp(P1:1994–2006))×100%

O3_netp(1994–2006) represents the average net O3 production for P1, and O3_netp(2007–2019) represents the average net O3 production for P2 for the specified four source regions over the free and lower troposphere, respectively.

To quantify the spatial influence of different source regions on low O3 levels over WNA, we calculated the normalized mean density (we call this as “relative intensity”). This value represents the mean SRR of air masses with positive net O3 production for a specific region, normalized against the hemispheric average (0–90° N) – SE Asia, NE Asia, Siberia, and the Tropical Pacific – during two time periods: P1 (1994–2006) and P2 (2007–2021). These values were derived from two-dimensional joint probability density functions (2D joint PDFs), estimated using a non-parametric, gaussian kernel density estimation (KDE) method, to estimate the PDF of a variable (Chen, 2017), implemented via the SciPy library in Python v3.7. Specifically, the relative intensity of region i during a given time was computed as (KDE (region i, time) / KDE (total, time)) × 100 (%), where region i ∈ {SE Asia, NE Asia, Siberia, Tropical Pacific}.

Results were also tested across multiple bandwidths (i.e., smoothing parameters for KDE). The optimal bandwidth values were determined using the leave-one-out cross-validation (LOOCV) method, also implemented via the SciPy library. This approach minimizes the estimated error of the KDE by iteratively computing the density estimator on all but one data point and then evaluating its ability to obtain the density of the left-out point, ultimately selecting the bandwidth that minimizes the total error across all data points. Uncertainty in the computed relative intensity for each sector was also assessed using bootstrap-estimated standard deviation (N = 1000).

Figure 2 illustrates the SRR maps for low (< 33rd percentile) and high (66th-95th percentile) O3 concentrations across the lower (0–3 km) and middle (3–8 km) troposphere. SRR maps are intrinsically linked to trans-Pacific maritime transport and are heavily influenced by a mixture of maritime pathways, continental air masses, and source signatures (Ryoo et al., 2017). From the map, we identify distinct transport corridors associated with varying O3 levels. In the lower troposphere (0–3 km), low O3 air masses exhibit high SRR values, with their peaks and dispersions aligned specifically with the southern regions of East Asia and the tropical Pacific (Fig. 2a and c); SRR is higher in the red and gray boxed regions on the map). Conversely, elevated O3 concentrations are primarily associated with air masses transiting through NE Asia (Fig. 2c).

This latitudinal contrast is further amplified in the free troposphere (3–8 km). While SRR values are generally larger in this layer compared to the lower troposphere, the spatial patterns remain consistent (Fig. 2b and d). For lower O3 percentiles, air parcels predominantly traverse the tropical eastern Pacific and SE Asia. In contrast, higher O3 percentiles are linked to air masses with extended SRR in the midlatitudes (15–45° N), exhibiting transport pathways that originate or pass through NE Asia. Similar patterns were reported by Cooper et al. (2010), who showed distinct transport features corresponding to different O3 levels, particularly in springtime.

These characteristic transport dynamics are also evident on a seasonal basis: transport patterns weaken slightly and become more localized over WNA during boreal summer (June–August; JJA), whereas they strengthen during boreal spring (March–May; MAM) and boreal winter (December–February; DJF) (Fig. S4).

This seasonal behavior aligns with findings by Ansari et al. (2025), who utilized explicit NOx and VOC emission tagging to demonstrate an increasing influence of foreign anthropogenic NOx and, to a lesser extent, global shipping NOx, over the northwestern United States between 2000 and 2018. Similarly, Li et al. (2023a) used the same explicit dual-tagging technique and found increasing contributions of Asian NOx emissions to Western US O3, particularly during winter from 1995 to 2019, further supporting our analysis. Taken together, these findings confirm that the geographic origin of an air mass is a key driver of O3 variability over WNA.

To examine transport trends on decadal scales, we analyze the SRR patterns for air parcels containing low free tropospheric O3 amounts over WNA. Figure 3 illustrates trends in the SRR of low-O3 air parcels from the lower and the free troposphere to the UTLS. For low O3, SRRs show no clear trend in the lower troposphere (Fig. 3a and b) or in the free troposphere (Fig. 3c) in the regions of interest. In contrast, low O3 SRR exhibits an increasing tendency across all regions in the UTLS, though with large variability (Fig. 3d and e).

SRR for high O3 (66th–95th percentile, Fig. 3e) observations over WNA exhibit more positive trends and large variability. For example, SRR from NE Asia for high O3, both in the lower troposphere and the free troposphere, shows a distinct increasing trend (Figs. 2c, d, 3e, and S3). Conversely, high-O3 SRR shows a decreasing trend at higher latitudes (e.g., Siberia), while over the Tropical Pacific it exhibits increasing trend with substantial variability (Fig. 3e).

An increasing trend in the UTLS SRR of air masses is evident in Fig. 3 (8–13 km, Fig. 3d and e), influencing both low and high O3 over WNA. Although the 8–13 km altitude range corresponds to the UTLS region and influences stratosphere-troposphere exchange (STE), it also intersects commercial flight corridors and regions of frequent lightning activity. Therefore, the upward SRR trend in the 8–13 km layer cannot be uniquely attributed to STE. Furthermore, we intentionally excluded extreme O₃ events from our primary analysis to mitigate STE influence. Investigating whether this increasing SRR trend points to a potential rise in STE activity requires further study and is beyond the scope of this work.

To further investigate the potential cause of increasing amounts of O3 in the lowest percentiles at 8–13 km, we examine aircraft emissions averaged over 1994–2019, as shown in Fig. 4a. Notably, aircraft emissions – which have increased steadily across source regions (Fig. 4b) – typically peak within the 8–13 km range, coinciding with the primary cruising altitudes of commercial aviation. This finding aligns with Eastham et al. (2024), who demonstrated that the impact of civil aviation is driven primarily by a hemispheric-scale tropospheric O3 response to NOx rather than by localized effects.

Beyond aviation emissions, we also find from GMI simulations that lightning-NOx shows increasing patterns over SE Asia and North America within the free troposphere and UTLS (3–13 km) (see Fig. 5 of Cui et al., 2025). These results further support the findings of Gressent et al. (2014), who identified additional sources – such as lightning-generated nitrogen oxides – as important contributors to O3 variability.

While Fig. 3 shows an upward trend in transport from the tropical Pacific UTLS, the rise in low O3 over WNA is not solely driven by aircraft emissions. In fact, aircraft emissions are quite low in this specific area (Fig. 4a). This observation instead suggests the influence of other emission sources. To investigate additional anthropogenic emissions over our source regions, we particularly examined shipping emissions. As shown in Fig. 5, shipping emissions are high over the Tropical Indian Ocean (including the SE Asia region) and the Tropical Pacific (Li et al., 2023b). While the decrease in global shipping NOx emissions around 2009 was likely not driven by a single factor, the observed NOx reduction in 2009 has been reported to be partially associated with the 2008 global economic recession (Tong et al., 2016; Fig. 5b).

During the recent period (P2), total shipping emissions show a slight increase over the tropical Indian Ocean and the midlatitude western Pacific (including Northeast Asia), with minor changes over the tropical Pacific. Relative to total anthropogenic emissions, shipping contributes up to 10.9 %, 17.7 %, and 95.2 % in these regions, respectively, highlighting the growing role of shipping emissions there (Fig. 5d).

Given that both transport and chemical processes can influence O3 levels over WNA, we further analyze trends in net O3 chemical production to distinguish the role of chemical processes using GMI. Figure 6 represents the net O3 chemical production obtained by subtracting GMI O3 loss from GMI O3 production across the NH and the four distinct regions (Fig. S1) in both the free troposphere and the lower troposphere. In the lower troposphere, net O3 production is typically an order of magnitude higher than in the free troposphere, with positive values occurring predominantly over densely populated regions of the NH, while net destruction is mainly observed over the tropical ocean (Fig. 6a and b).

To further assess temporal changes in net O3 production in the lower and free troposphere, we calculate the change in net O3 production between the early period (P1: 1994–2006) and the later period (P2: 2007–2019), as shown in Fig. 6e and f. The zonal-mean latitude–height cross sections of net O3 production during P1 and P2 reveal several key features: (i) a consistent positive net O3 production over 20–35° N in the lower troposphere during both periods (Figs. S5 and S6); (ii) a notable increase in net O3 production in the extended upper tropospheric and lower stratospheric region in recent years (pink color in Fig. 6e centered around 20° N and around 300–150 hPa) (P2, Fig. S6), along with a decrease in net O3 production over the tropical Pacific (green color extending up from the surface, centered on the equator). These results are broadly consistent with Archibald et al. (2020), who reported O3 production from the UK Chemistry and Aerosol Community Climate Model (UKCA; Luhar et al., 2018) for the year 2000. However, slight differences in the magnitude and pattern of extratropical zonal-mean net O3 production may stem from our study's longer analysis period (1994–2019) and the use of a different model. During the later period (P2), a significant increase and positive trend in ΔO3_netp (net O3 production difference) is observed, particularly at the low-tropospheric level over NE and SE Asia (Fig. S5), with the rate reaching up to ∼ 157 % over SE Asia (Fig. 6f).

O3 net production and destruction vary significantly by region (Fig. 6a and b), influenced by a combination of natural and human-caused factors. A study led by Thorp et al. (2021), for instance, found that in western Siberia, surface O3 is controlled by a dynamic interplay of seasonal atmospheric transport patterns, a dominant sink from dry deposition by forest vegetation, and the prevalence of anthropogenic emissions from the transport and energy sectors. However, note that this is the net O3 production in a particular model grid cell, and does not account for transport from the stratosphere, surface deposition or transport of O3 from one region to another.

Overall, Fig. 6 suggests that chemical ΔO3_netp over East Asia has shown a significant increase in recent years, particularly over SE Asia in the lower troposphere. To investigate the temporal evolution of O3 increases in more extended vertical layers, Fig. 7 illustrates the trends for O3 across the three vertical layers. In the tropics, tropospheric O3 exhibits an overall increasing trend, particularly over East Asia, consistent with the findings of Ziemke et al. (2019).

In the UTLS (Fig. 7a), O3 growth rates tend to increase at high latitudes, partially reflecting the influence of stratospheric processes. Conversely, large variability is observed in the extratropics due to a complex interplay between dynamic processes and O3 chemistry (Cooper et al., 2004; Bak et al., 2025). The pattern is different in the free troposphere with positive trends in O3 across most of the NH. Strongest increases stretch from South and East Asia across the North Pacific Ocean to WNA, while decreases are limited to the equatorial region (Fig. 7b). The increase in O3 growth rate in the free troposphere and the decrease in the lower troposphere shown in Fig. 7 are overall consistent with and supported by observed O3 trends (Chang et al., 2023a; Fig. S1). In addition, O3 shows increases over SE Asia and the Pacific Ocean but decreases over North America, aligning with IAGOS dataset trends reported by Gaudel et al. (2020). Similar patterns are also found in net O3 production trends (Figs. S5–S7).

The schematics in Fig. 8 summarize our findings on the key regions contributing to low O3 over WNA. Regional net O3 production analysis indicates that SE Asia's contribution to low O3 (< 33rd percentile) levels over WNA – driven by both SRR and chemical production – has intensified in recent years (P2: 2007–2021) compared to the earlier period (P1: 1994–2006). This increase is most pronounced in the lower troposphere, where relative O3 production rose by approximately 23 % (Fig. 8b), with the highest positive net production centered over SE and NE Asia (Fig. S6). While net O3 production from SE Asia also increased within the free troposphere, the magnitude of change was less substantial (Fig. 8b). In contrast, the UTLS exhibits positive net O3 production across nearly all regions. This widespread increase is linked to rising aircraft emissions (Fig. 4) and enhanced SRR (Fig. 3) – a result consistent with the hemispheric-scale response described by Eastham et al. (2024). Notably, the tropical Pacific also shows entirely positive net O3 production within this upper layer.