The largest and most persistent variability in tropospheric ozone originates from seasonal cycles due to seasonally varying stratospheric ozone influx and photochemical production in the troposphere. Seasonal cycles for EPIC TCO were determined from daily data by first computing the 2015–2024 average for each day of the year. The daily climatology was then averaged over 3-month seasons to produce the seasonal climatological maps shown in Figure 5. The seasonal cycle panels in Figure 5 for EPIC TCO can be compared with Supplementary Figure S1, S2 for OMI and OMPS TCO (discussed below).

NH TCO in Figure 5 is the greatest from spring (March–April–May, MAM) to summer (June–July–August, JJA), with largest ozone level occurring in the Mediterranean region during JJA. The Mediterranean in summer is known as an annually recurring accumulation “crossroads” region for pollution, including ozone precursors (Lelieveld et al., 2002). Terrain height is seen to reduce regional TCO by as much as ∼10–15 DU, such as over the Rocky Mountains in western N. America, the Asian Tibetan Plateau, and the S. American Andes. In the tropics and subtropics, the dominant signature in TCO is a zonal maximum in the Atlantic and minimum in the Pacific (i.e., zonal “wave one”). This wave-one pattern was first documented by Fishman et al. (1990). The wave-one in TCO is present year-round, with the largest amounts of ∼45 DU in the Atlantic in SON. Differences in TCO between the Atlantic and Pacific are also greatest during the SON (September–October–November) season, which is up to ∼25 DU. The structure and time dependence of the wave-one is due to the persistent tropical east–west Walker circulation, which controls the transport of ozone produced from STE, and ozone precursors produced by lightning and biomass burning into the tropical Atlantic Ocean region from Africa and S. America during the peak burning months (September–October) each year.

As further validation of EPIC TCO, we computed a zonal mean EPIC TCO climatology for each month and compared it with similar OMPS and OMI zonal mean TCO monthly climatologies averaged over the same time period of 2015–2024 (Figure 6). The Global Modeling Initiative (GMI) replay 12-month climatology is also included in Figure 6 for comparison. The GMI simulation climatology uses MERRA-2 meteorology and the same MERRA-2 tropopause with TCO derived by integrating model profile ozone from the ground up to the tropopause (Ziemke et al., 2015, and references therein).

In the NH, in Figure 6, there is a distinct propagating phase shift of the seasonal peak from April to May in the subtropics to June to July at mid-latitudes for all three products and model. That is, in each case, the seasonal maximum in TCO exceeds 40 DU in April–May at approximately 25^oN, which extends into June–July at approximately 40^oN. This NH phase transition corresponds with a northward rate of approximately 5^o of latitude per month. The lowest TCO in the NH extra-tropics occurs in October–February at 50^oN–60^oN for all three products. In the SH, in Figure 6, the TCO seasonal cycle has the same phase with maxima during September–November at all latitudes for all three products and model; the largest SH TCO occurs at 15^oS–20^oS in September–November, with the amount being ∼36 DU. The lowest TCO in the SH is ∼15 DU and occurs at high latitudes around spring–summer months. The model simulation at high SH latitudes has a seasonal cycle close to that of EPIC and OMI and suggests that there may be some outlier measurements present for OMPS TCO at the high SH latitudes in winter when solar zenith angles for the satellite measurements are the largest. There are differences between the products and model in Figure 6 that involve mean bias differences and algorithm issues, but the patterns and amplitudes are all closely comparable between all four TCO panels. We summarize the seasonal variability in the tropics by noting the strong dependence on hemisphere, with TCO peaking in March–April in the NH and in September–November in the SH.

Tropical ENSO events and ensuing signatures on the ocean and atmosphere have been extensively studied and documented. Trenberth (1997) provided an extensive overview of ENSO with many key references. ENSO events in the tropics are known to generate substantial inter-annual anomalies in tropospheric ozone from satellites, sondes, and models (Chandra et al., 1998; 2002; Fujiwara et al., 1999; Thompson et al., 2001; Sudo and Takahashi, 2001; Nasser et al., 2009; Lee et al., 2010; Oman, et al., 2013; Ziemke and Chandra, 2003b; 2015; Le et al., 2024; Bruckner et al., 2024). Large ENSO-related increases in TCO reaching +20 DU over Indonesia in October 1997 have been recorded by Chandra et al. (2002) in ozonesondes, the Geostationary Earth Observation System Chemical transport model (GEOS-Chem), and Total Ozone Mapping Spectrometer (TOMS) satellite measurements. The GEOS-Chem model included in Chandra et al. (2002) indicated that most of the increases in TCO in the western Pacific during the 1997 El Nino occurred over Indonesia and originated from biomass burning, with the largest biomass burning contribution coming from above 5 km altitude.

We have studied inter-annual changes of EPIC TCO for 2015–2024 using de-seasonalized monthly-mean measurements. De-seasonalization of TCO was accomplished by subtracting mean seasonal cycles calculated from the 10-year record. Figure 7a compares time series of de-seasonalized TCO from EPIC (blue), OMI (red), and OMPS (black) for 2012–2024 in the tropical western Pacific at 100^oE and 5^oS, where the ENSO signal is large (discussed in Section 3.2.3). There were two strong El Nino events during this 10-year record, one in 2015 and one in 2023, with corresponding increases of 5–10 DU for all three products. La Nina events in late 2017 and 2020, seen as negative values in the Nino 3.4 index (Figure 7b), correspond with negative TCO anomalies of approximately −5 DU in the western Pacific.

Another source of tropospheric ozone variability comes from the Indian Ocean Dipole (IOD) (Saji et al., 1999). A positive-phase IOD corresponds to enhanced convection just east of Africa, with suppressed convection toward the east over the Indian Ocean. The positive IOD in 2019 was among one of the most intense ever recorded and created drought conditions over northern Australia due to the induced lack of convection and rainfall (Lu and Ren, 2020). The increase in the IOD index (Figure 7b) corresponds to a positive TCO anomaly of approximately +5 DU in late 2019 due to suppressed convection over the tropical western Pacific region for both OMI and OMPS (note that there is a gap in EPIC data between late June 2019 and February 2020).

The black arrows in Figure 7 indicate the three largest (positive) TCO anomalies of +5 to +10 DU during this time period. These large positive anomalies in TCO coincide when the ENSO and IOD indices are both in a generally positive phase.

We further study variabilities in EPIC TCO using a MLR model (e.g., Randel and Cobb, 1994, and references therein) to remove the dominant seasonal cycle and ENSO forcings to reveal other remaining variabilities. Prior to this MLR analysis, we re-gridded the EPIC daily TCO from 1^o × 1^o to 5^o × 5^o bin size to reduce computational time.

Table 1 shows the calculated correlations between the three TCO products and the individual ENSO and IOD indices from Figure 7. Correlations between IOD and TCO for all three products are small compared to correlations between ENSO and TCO. These small correlations coupled with redundancy between the IOD and ENSO indices (their cross-correlation is +0.47) suggests that we include only ENSO and not IOD as TCO proxies in our MLR analysis.

The MLR model that we use is T C O t = A t + B · t + C · E N S O t + R t , where t is the day index that runs between 1 January 2015 (t = 0) and 31 December 2024 (t = 3,653), A(t) is the seasonal-cycle fit, B is the linear trend coefficient, C is the ENSO coefficient, with ENSO(t) being the Nino 3.4 monthly time series extrapolated to daily, and R(t) represents the residual error term for the model. The seasonal cycle A(t) uses nine regression constants (a₀, a₁, …, a₈): A(t) = a₀ + a₁ cos(ωt) + a₂ sin(ωt) + a₃ cos(2ωt) + a₄ sin(2ωt) + a₅ cos(3ωt) + a₆ sin(3ωt) + a₇ cos(4ωt) + a₈ sin(4ωt), where the angular frequency ω = 2π/365. The other coefficients in the MLR model are simple constants. Any missing TCO data (such as during the earlier half of 2015 for EPIC) are not included in the regression. We include a linear trend term in the model, even though the record length for EPIC is too short to derive reliable trends.

Figure 8 shows a breakdown of the MLR model for each coefficient term including the original data TCO(t) time average (Figure 8a) and standard deviation of the MLR residual R(t) (Figure 8e). Figure 8a shows the 10-year average original data TCO, which can be compared with the 10-year seasonal climatology plotted in Figure 5. Figure 8b shows that the MLR model seasonal cycle compares closely with the original TCO in Figure 8a.

The linear trend term in Figure 8c suggests small increases in TCO in the tropics, with decreases in the extra-tropics of both hemispheres. The decrease in NH and SH TCO during 2015–2024 is consistent with anomalous drops in TCO of a few DU in both hemispheres starting in spring 2020 during the COVID-19 pandemic (Figure 1). Trends in global TCO have been reported in several articles published for the Tropospheric Ozone Assessment Report (TOAR-I and TOAR-II; https://igacproject.org/activities/TOAR/), where trends exhibit large variations in patterns and large differences between using different products. A main result from TOAR-II for the satellite, ground, and aircraft measurements is that TCO has been increasing over the East Asian region for several decades, with numbers varying from +1 to +4 DU decade⁻¹. TOAR-II focused on trends ending in 2019, while in this paper we consider 2015-2025. Most EPIC TCO trends in Figure 8c are not positive, but instead zero to negative and can be attributed to the short EPIC record coupled with effects from the 2-3 DU drops in NH TCO during COVID-19.

The ENSO coefficient in Figure 8d shows that ENSO variability in TCO is largely a tropical signal between the eastern and western Pacific. However, the ENSO signal in Figure 8d is not statistically zero outside the tropics. It is well known that the tropical ENSO affects meteorology in the extra-tropics (Vimont et al., 2003; Sun et al., 2021). The residual standard deviations in Figure 8e shows that the MLR model does well in simulating TCO in the tropics and subtropics with values of ∼2–4 DU (∼10% of background), but it becomes larger in higher latitudes, especially in high northern latitudes where numbers are closer to 5–6 DU (∼15%).

An additional way to evaluate the MLR results is to compare the spatial variability patterns for each model component. To do this, we compute the standard deviation of the reconstructed 10-year daily time series (t = 0–3,653 days) at each location for each term in the regression model, as shown in Figures 9b–e and compare to the standard deviation of the original TCO time series shown in Figure 9. Notably, the panels in Figure 9 have varying color bar scales. Figures 9a, b show the standard deviations for the original data and MLR seasonal cycle fit, respectively. Figures 9a, b show that the largest standard deviations for annual changes in TCO (up to ∼8 DU) occur over landmasses and along storm-track regions just east of the N. American and Asian continents. In comparison, the trend and ENSO terms have smaller standard deviations with trends showing wave structure outside the tropics, and ENSO again is associated mostly with the tropical Pacific region. ENSO standard deviations multiplied by a factor 4 are ∼5 DU, which results in approximate peak-to-peak average change from ENSO.

The standard deviations for MLR residuals in Figure 9e (same data as plotted in Figure 8e) are smaller than the original data variabilities (Figure 9a) and seasonal-cycle fit (Figure 9b) but are still up to 5 DU or greater in mid-high latitudes. The residuals show us that there are large variabilities present in daily TCO aside from just the seasonal cycles and ENSO forcing. Our further investigation of TCO variability uses the calculated residuals R(t) from the MLR model. The analysis includes both extra-tropical weather systems and the tropical 1–2 month MJO. We provide only a short overview of the mean amplitudes and geographical patterns associated with these forcings by applying bandpass filtering (Stanford and Vardeman, 1994) to the EPIC TCO 10-year record of daily residuals R(t) for 4–10 day and 30–60 day periods (Supplementary Figure S4). A detailed investigation of baroclinic weather systems and the MJO in tropospheric ozone is beyond the scope of our study.

Weather systems in both hemispheres are present year-round in atmospheric temperature, pressure, winds, and trace gases with characteristic oscillation periods of ∼4–10 days and large planetary scales involving east–west zonal wavenumbers of ∼4–10 (Andrews et al., 1987). We have estimated the average peak-to-peak changes in EPIC TCO associated with weather systems by first calculating standard deviation from bandpass filtered data and then scaling by a factor of 4 to approximate peak-to-peak change. Figure 10a shows a map of these changes for EPIC TCO associated with weather systems with 4–10-day periods. Figure 10a shows that TCO variations up to 10 DU occur throughout much of the SH and also over northern America and eastern Asia. The largest anomalies in TCO associated with weather systems have persistent patterns that are associated with natural topography, including the storm-track regions east of the continents of N. America and Asia. The percent changes in Figure 10c show that relative changes in the SH extra-tropical TCO due to weather systems may be up to 40% of the background amounts.

Intra-seasonal variability in the tropics is dominated by the 1–2 month MJO, known historically as the 40–50 day MJO (Madden and Julian, 1972; 1971). Madden and Julian (1994) provide a detailed overview of MJO effects on the ocean and atmosphere. The MJO is associated with anomalous deep convection in the tropics driven by ocean–atmosphere feedback coupling with convection cells starting in the Indian Ocean and propagating eastward across the dateline into the eastern Pacific with periods averaging approximately 45 days. The largest MJO signal in meteorological parameters (winds, temperatures, and pressures) occurs during northern winter months; however, this oscillation is present year-round.

Figure 10b shows bandpass-filtered EPIC TCO for the period of 30–60 days to estimate peak-to-peak amplitudes in the tropics and subtropics associated with the MJO. The 1–2-month signals in TCO in Figure 10b are generally tropical compared to patterns associated with weather systems (Figure 10a), and they also have a smaller magnitude of variations of ∼5 DU. At high NH latitudes, in Figure 10b, the signal is up to ∼4 DU. Studies have shown that the MJO impacts dynamical events including blocking in NH high latitudes (Henderson and Maloney, 2018, and references therein), so we cannot rule out that the observed TCO anomalies in high NH latitudes (Figure 10b) are related to MJO. Figure 10d shows associated percentage changes relative to mean background ozone. Large percentage changes that occur in high southern latitudes in Figure 10d are due to small background TCO amounts year-round.