Since 2015, the Institute of Environmental Physics (IUP), University of Bremen, and Alfred-Wegener-Institut (AWI), Potsdam, have conducted atmospheric observations at the Palau Atmospheric Observatory (PAO) in Koror, Palau (7.3° N, 134.5° E) . PAO is located on the Palau Community College (PCC) campus. The O3 concentration was measured by solar absorption Fourier Transform infrared (FTIR) spectrometry in September and October 2022. The information of the FTIR measurement are given in Table . FTIR measurements have been performed in one of the PAO scientific containers using a Bruker IFS 120M spectrometer equipped with indium antimonide (InSb) since 2018. This allows to record spectra from 1900 up to 6000 cm-1. In August 2022, Mercury Cadmium Telluride (MCT) was installed to cover the spectral region between 700–3000 cm-1. The lower wavenumber region allows for the study of concentration profiles with higher precision, which is especially important for ozone.

The FTIR spectra are recorded at a high spectral resolution up to 0.005 cm-1 by pointing the solar tracker at the sun during cloud-free weather conditions. The weather conditions can be actively monitored by the sky camera mounted above the window of the laboratory container to minimize the influence of clouds on solar absorption FTIR. We performed measurements throughout the day to record the spectrum from sunrise to sunset. The number of FTIR measurements contributing to each hourly bin ranges from 10 to 55, with the highest sampling between 11:00 and 14:00 LT. Each hourly bin includes measurements from multiple independent days (typically 3–10 dh-1, see Appendix ). The spectra collected after 17:00 LT were excluded from our analysis due to shading from a tree adjacent to the laboratory container. The measurements during the campaign period were grouped by hour and averaged to evaluate the diurnal cycle of O3. Therefore, we successfully obtained the ozone daytime diurnal cycle from 07:00 to 16:00 LT.

The retrieval of trace gas concentrations from solar absorption FTIR spectra was performed using SFIT-4 (Spectra Least Squares Fitting) software. Spectral line parameters were taken from the high-resolution transmission molecular absorption database version 2020 (HITRAN2020) . A priori profiles were kept constant for the campaign and were from the Whole Atmosphere Community Climate Model (WACCM V4). O3 was retrieved in three microwindows: 1000.00–1000.08, 1001.00–1001.30, and 1003.16–1004.50 cm-1 with a simultaneous fit of H2O and other interfering species (Table. ), retrieval strategy is consistent with the the Network for the Detection of Atmospheric Composition Change (NDACC) framwork .

To quantify tropospheric ozone, we use the part of the retrieval corresponding to the tropospheric degree of freedom (DOF≈1), whose associated partial column averaging kernel (PC AVK) spans the vertical range from the surface to 10.2 km (see Fig. a, red curve). The vertical range we use in this study represents the low troposphere in the tropical region. This layer is used to calculate the tropospheric dry-air partial column–averaged mole fractions of ozone, XO3,p. The tropospheric XO3,p is calculated as: 1XO3,p=PCO3,pPCdryair,p=PCO3,pPCwetair,p-PCH2O,p, where PCO3,p, PCwetair,p, and PCH2O,p are the partial columns (in moleculescm-2) of ozone, wet air, and water vapor, respectively, over the vertical range p=0.2–10.2 km. XO3,p represents the dry-air partial column-averaged mole fraction of ozone, which is equivalent to a column-weighted mean mixing ratio (ppb) calculated by dividing the ozone partial column by that of the dry air. This approach is adopted to facilitate direct comparison with ozonesonde profiles and model outputs. For simplicity, XO3,p is hereafter referred to as the tropospheric ozone column-averaged mole fraction (TOC in ppb), over 0.2–10.2 km. Noting that its unit (ppb) distinguishes it from the total column abundance expressed in Dobson Units (DU). The altitude range is consistent with previous FTIR studies defining TOC over 0–10 km e.g..

The uncertainty budget of our ozone retrievals includes different types of contributions as shown in Table . For the total column, systematic uncertainties from spectroscopy and instrument modeling amount to about 5.5 %, while random noise contributes 1.6 %. For the tropospheric column, the estimated uncertainties are of a different nature and therefore not directly additive. Day-to-day variability (±3.4 ppb) reflects the representativeness of the retrievals.

When interpreting the diurnal pattern from FTIR-derived TOC, two aspects require consideration. First, the FTIR-derived TOC is the integration over 0.2–10.2 km, and this inevitably dampens near-surface variability, assuming there is little variability of ozone in the free troposphere over Palau in a pristine oceanic region, like the TWP. This compromises the representation of boundary-layer changes in the integrated TOC. Second, besides this vertical smoothing, other sources of uncertainty also affect the retrieved TOC, such as solar zenith angle (SZA) dependence and residual stratospheric influence.

As shown in Fig. a, the PC AVK of the first layer is highly sensitive within the 0.2–10.2 km range, indicating high information content. This supports using the part of the retrieval corresponding to the tropospheric degree of freedom (DOF≈1) to represent the TOC. Moreover, the PC AVK is the highest below 2 km (PC AVK ≈1.0), and decreases with altitude (Fig. a, red line). This indicates that TOC mainly reflects near-surface information. The same feature was observed for all PC AVK during the measurement period (Fig. b). The maximum sensitivity always lies near the surface, where most of the diurnal variation occurs. From previous aircraft observations, no discernible diurnal ozone variations were found above 750 hPa . Thus, consistent with the PC AVK, the FTIR-derived tropospheric ozone column primarily captures near-surface ozone variability, although vertical integration over the column reduces the apparent diurnal amplitude. We estimate that the TOC retains only about 40 % of the near-surface variability (see Appendix ).

From FTIR-derived TOC to interpret the diurnal cycle, the SZA dependency of the measurements should also be considered. As shown in Fig. b, the sensitivity of the FTIR measurements varies with SZA. We estimate the resulting SZA-induced uncertainty of a maximum deviation of 1.9 ppb (7.6 %), see details of the quantifying method in Appendix . To validate the FTIR TOC diurnal cycle, we further compare the retrievals with model simulations and ozonesonde observations.

However, the PC AVK does not drop sharply above 10.2 km but gradually decays, reaching near-zero only around 20 km (Fig. ). This indicates residual sensitivity above the defined TOC layer, which can lead to vertical smoothing–induced leakage of stratospheric ozone to the TOC. Given that ozone mixing ratios increase with altitude from the upper troposphere, this leakage results in an overestimation bias of approximately 1.5 ppb (6.0 %) in the retrieved TOC, see Appendix . Together, these values characterize the expected range and type of uncertainty, but they do not compromise the detection of the observed diurnal pattern.

O3 concentrations were simulated using the global 3-D chemical transport model GEOS-Chem (version 13.4.0, classic, , 2022; , 2001), with the full-chemistry module. The model was driven by meteorological fields from the Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2), provided by the NASA Global Modeling and Assimilation Office (GMAO).

Emissions were handled using the Harmonized Emissions Component (HEMCO; ). Lightning NOx (LNOx) emissions were taken from the GEOS-Chem standard configuration (HEMCO Lightning NOx v2014-07), based on cloud-top height parameterizations and described in the official GEOS-Chem HEMCO archive.

Moreover, we conducted two sensitivity simulations, one LNOx emission is turning off and one LNOx is doubling, to show lightning effect on ozone formation and atmospheric oxidation. Simulations cover the period 2020–2022, using a horizontal resolution of 2°×2.5°, 72 vertical levels from the surface to 10 hPa, with time steps of 10 min for transport and 20 min for chemistry. Model output was archived hourly.

O3 concentrations from FTIR measurements, ozonesonde profiles, and model simulations are compared in this study to evaluate the consistency between observations and model outputs. To ensure a fair comparison, we use model results from the full chemistry simulation and account for the vertical sensitivity of the FTIR retrievals by applying the retrieval AVK.

The retrieval sensitivity is described by the AVK matrix A and the a priori profile xa, both of which affect how true atmospheric profiles are represented in the FTIR product . To make the model or ozonesonde profiles xm comparable with the FTIR retrievals, they are first smoothed using the following equation: 2xs=xa+A(xm-xa), where xs denotes the smoothed profile that incorporates the FTIR sensitivity characteristics. To ensure consistency in the application of the AVK, all model and ozonesonde profiles were interpolated to the vertical grid of the FTIR retrieval. This step is essential, as the AVK matrix A is defined on the FTIR retrieval levels, and applying it directly to profiles on different vertical coordinates would lead to incorrect smoothing. Linear interpolation was used in pressure space, as it preserves the structure of atmospheric layers and is commonly applied in intercomparison studies e.g. .

After smoothing, the dry-air partial column-averaged mole fraction of ozone, XO3,p, is computed for both the FTIR retrievals and the model or ozonesonde profiles as mentioned in Sect. Eq. (). The partial column is defined from the surface up to 10.2 km, corresponding to the first layer of the FTIR retrieval product, as described in Sect. . This consistent vertical integration ensures that differences reflect physical and chemical discrepancies rather than vertical resolution mismatches.

To investigate the large-scale transport influencing ozone variability over Palau, we performed air-parcel trajectory analyses using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model developed by NOAA's Air Resources Laboratory . The calculations were driven by meteorological fields from the Global Data Assimilation System GDAS;, which provides global coverage at 1°×1° horizontal resolution and 3-hourly temporal frequency. GDAS data are widely used in regional and long-range transport studies.

Backward trajectories were initialized at the location of the Palau Atmospheric Observatory for three starting altitudes: the surface, 5, and 10 km. These levels represent conditions within the marine boundary layer, the lower free troposphere, and the upper portion of the FTIR retrieval range, respectively. Each trajectory was traced 10 d backward in time to identify the dominant pathways and potential source regions influencing the observed ozone during the FTIR campaign.

The resulting trajectory ensemble provides a dynamical context for interpreting the tropospheric ozone signal by indicating whether the sampled air masses originated from remote oceanic regions, convective outflow, or areas with continental influence. While GDAS does not fully resolve small-scale boundary-layer mixing, it reliably captures the larger-scale circulation patterns that dominate transport in the TWP, making it suitable for assessing the origin and history of air masses arriving at Palau.

We present tropospheric ozone observations over Palau using FTIR spectroscopy. Figure  shows the time series of daily mean TOC derived from FTIR measurements, compared with GEOS-Chem model simulations and ozonesonde observations. The FTIR observations reveal very low tropospheric ozone levels, with an overall daily mean of 24 ppb. GEOS-Chem simulations, smoothed with the same AVK, yield a mean of 22.85 ppb over the campaign period. The ozonesonde-based TOC shows good agreement with the FTIR measurements, further validating the reliability of the retrievals. As shown in Fig. b, the differences between the model and FTIR observations generally vary within ±5 ppb, indicating that the model captures the day-to-day variability reasonably well. Overall, the model underestimates the absolute concentrations.

Building upon the time series analysis presented above, we next focus on the daytime diurnal variability of tropospheric ozone derived from the high-temporal-resolution FTIR observations. Because of the continuous solar absorption measurements throughout the day, the FTIR provides hourly retrievals between 07:00 and 16:00 LT. These retrievals can be used to derive the diurnal pattern of the O3 measurements, as described in Sect. . Figure  summarizes the diurnal behaviour: panel a shows the FTIR-derived tropospheric ozone column (TOC) together with GEOS-Chem simulations, while panel b presents normalized ozonesonde ozone profiles at different hours for evaluating the FTIR-derived diurnal pattern.

Figure a shows the daytime diurnal cycle of tropospheric ozone from FTIR measurements compared with GEOS-Chem simulations. FTIR data show an increase in ozone in the early morning, peaking around noon, followed by a decline in the afternoon. Figure  also compares the FTIR diurnal cycle with the GEOS-Chem simulation, time matched to FTIR measurements and smoothed. Compared with observational data, the model shows a much flatter pattern. Simulations without applying AVK smoothing display an even flatter diurnal variation with ozone concentrations remaining nearly constant throughout the day (see Appendix ). Note that Fig. a shows TOC derived from model profiles after AVK smoothing. The weak midday enhancement of about 2 ppb mainly results from the AVK smoothing effect associated with changes in solar zenith angle during the day. This indicates that the model fails to capture both the midday peak and the amplitude of the observed variations simultaneously. It also overestimates ozone in the early morning and late afternoon. Consequently, the simulated ozone diurnal cycle appears muted compared to observations. While the model captures the characteristic diurnal variability of photochemical radicals, the resulting net ozone change is insufficient to drive a discernible amplitude in the column (see Appendix  and Fig.  for details on radical concentrations). This discrepancy likely arises from the coarse horizontal resolution and parameterized boundary-layer dynamics, which cannot catch the localized mixing or production mechanisms in the TWP. In addition, simplified representations of diurnal emissions and photolysis may further damp small-scale variability, as suggested by the FTIR and ozonesonde diurnal pattern in Fig. b.

Figure b compares normalized ozone variations from FTIR and ozonesonde measurements (see Sect. ). Because the FTIR retrievals have different sensitivity across altitude, the AVK must be considered when making such comparisons . Given the limited number of time-matched observations, we used normalized hourly averages to remove absolute differences between datasets, see Sect. . This approach enables a qualitative comparison of diurnal variability based on relative changes without applying smoothing to the ozone sounding profiles. Relative deviations from the mean were calculated for each dataset (Eq. ), and the resulting normalized diurnal variation is shown in Fig. b. The absolute hourly mean plot of ozonesonde is shown in Appendix . The ozonesonde data also exhibit a midday enhancement, though less pronounced than in the FTIR measurements, followed by a discernible afternoon decline and a more dispersed distribution in both the near-surface layer (below 2.5 km) and the free troposphere (2.5–10 km). These results highlight the inherent challenges in comparing FTIR and ozonesonde observations due to their different vertical sensitivities. Even without sufficient morning ozonesonde launches and without AVK smoothing of profiles, the normalized O3 from different datasets still provides indirect evidence for the pattern of midday peak and the afternoon decline. The error bars in Fig. a indicate an hourly variability of approximately 2–6 ppb (±1σ). The daytime diurnal variation of tropospheric ozone measured by FTIR is on the order of 6–8 ppb, but this magnitude should be interpreted as a lower limit of near-surface ozone variability, as the FTIR retrieval represents a vertically integrated tropospheric column (0–10 km) with a reduced peak-to-peak amplitude.

This interpretation is supported by the ozonesonde observations in Fig. b, which show larger ozone variability near the surface than in the free troposphere, where photochemical production and dry deposition are weaker above approximately 750 hPa (about 2.5 km) . Accounting for the solar zenith angle (SZA) effect, which leads to an overestimation of the apparent diurnal amplitude by about 1.9 ppb (Appendix ), the FTIR-derived peak-to-peak variation is approximately 4–6 ppb. Considering the ∼40 % damping of near-surface variability in the vertically integrated tropospheric ozone column (Appendix ), this implies a near-surface diurnal ozone amplitude on the order of ∼15 ppb.

Palau is located in the TWP, a region characterized by persistently low tropospheric ozone concentrations . Observations show that daytime ozone levels typically remain between 10–30 ppb from July until October , among the lowest values globally. To examine the origin and transport pathway of air masses influencing Palau, we computed 10 d backward trajectories using the HYSPLIT model. Most trajectories remain below 10 km (Fig. a), indicating that transport occurs within the free troposphere. Trajectories longer than 8 d predominantly originate from the central Pacific, following easterly circulation across the Pacific, with trade winds dominating the lower troposphere and additional contributions from large-scale tropical circulation at higher altitudes (Fig. b). This flow pattern is typical for the September–October period, which lies in the transition between the southwest monsoon and the establishment of the northeast trade wind regime . During this period, the TWP is mainly influenced by persistent marine inflow from the east, resulting in minimal continental influence. Ozone lifetimes are on the order of 10–20 d in the free troposphere , but considerably shorter in the marine boundary layer and increasing with altitude. (about 5 d; ). Thus, these air parcels are expected to retain their low ozone concentrations upon arrival. In addition, backward trajectories show that during the 2–6 d before arrival, air masses confined mostly below 1.5 km originate in the western Pacific, reflecting additional regional marine contributions (Fig. a and b). Together, these results suggest that both long-range and regional transport of ozone-poor air masses contribute to the persistently low tropospheric ozone observed over Palau.

Satellite retrievals support this interpretation. A broad ozone minimum is evident over the western Pacific warm pool (Fig. c), coinciding with low column densities of major precursors: CO (Fig. d), HCHO (Fig. e), and NO2 (Fig. f). The scarcity of these precursors in both the lower and free troposphere points to a suppressed photochemical ozone production regime. The lack of precursors arises from weak local emissions and the continuous inflow of clean marine air from the eastern Pacific. In addition, the interhemispheric convective zone (ITCZ) over the TWP during the campaign coincided with strong precipitation bands , which likely enhanced the washout of precursors. Enhanced humidity in the tropical troposphere promotes ozone loss via OH chemistry. At the same time, precipitation efficiently removes soluble species such as HCHO and nitrogen reservoir species produced from NO_x oxidation (e.g. HNO3), indirectly reducing NO_x availability and suppressing ozone production. Trajectory analyses confirm that air masses reaching Palau predominantly follow oceanic pathways with minimal continental influence, as mentioned before. This is consistent with , who showed similar oceanic transport patterns in the 5–10 km layer but predominant during September and October. These pathways align with regions of consistently low CO and ozone concentrations across the tropical Pacific (Fig. c and d). In contrast, NO2 has a short atmospheric lifetime and is not efficiently transported over long distances, such that its low abundance near Palau reflects weak local and regional NOx sources (Fig. f). HCHO, while also short-lived, is primarily produced locally through VOC oxidation, and its low abundance therefore indicates weak photochemical activity under pristine marine conditions (Fig. e), further limiting in situ ozone formation. The low NO2 columns also suggest minimal contributions from lightning or other episodic NOx sources, reinforcing the interpretation of limited photochemical ozone production.

In the maritime TWP, despite low lightning frequency, lightning-generated NOx remains an important free-tropospheric ozone source. We therefore investigate the microphysical conditions influencing lightning initiation, in addition to dynamical and chemical factors. Figure a shows the spatial distribution of lightning stroke density from WWLLN observations, confirming that lightning activity near Palau is lower than over the Maritime Continent or northern Australia. Although isolated lightning events do occur over the open ocean (Fig. a), they are sparse and less frequent, indicating that meteorological and microphysical environments in this region are inherently unfavorable for intense convective electrification. Observational evidence for marine convection without lightning has been reported in several studies, including over the tropical Atlantic and the PEM-West campaign , which found that such convection favors the washout of NOx derivatives.

The MODIS-derived cloud effective radius is shown in Fig. b. Over the warm pool region, cloud droplets are generally larger, which is typically associated with enhanced precipitation water content and reduced ice water content . These conditions promote warm-rain processes and inhibit the development of ice particles, thereby weakening charge separation and suppressing lightning activity . Figure c displays the dust aerosol optical thickness over the TWP, showing low dust loadings above Palau. In contrast to continental outflow regions, the central Pacific air column is nearly devoid of aerosol particles capable of serving as ice nuclei. This lack of ice nuclei suppresses the formation of ice-dominant or mixed-phase clouds , reducing the likelihood of charge separation and lightning initiation . Low aerosol concentrations favor the formation of larger cloud droplets, which reduces lightning activity and lowers NOx production. This, in turn, diminishes ozone levels in the upper free troposphere and illustrates the tight coupling between aerosol microphysics, cloud dynamics, and atmospheric chemistry. It provides an additional causal chain rooted in microphysical processes, through which precursor scarcity is further reinforced, complementing the direct explanations of weak anthropogenic influence and efficient washout (Sect. ). In this view, the low precursor abundances over Palau arise not only from a dynamically clean marine environment but also from suppressed lightning NOx production linked to the paucity of ice nuclei and mixed-phase clouds.

While this interpretation is supported by the observed patterns, the spatial differences (Fig. ) between dust aerosol, cloud effective radius, and lightning activity suggest that more complex microphysical processes may also be at play. This points to the possibility of additional factors influencing lightning production in marine convective systems, such as variations in updraft strength, cloud ice content, freezing level height, or the availability of giant cloud condensation nuclei for mixed-phase cloud . Importantly, the overall meteorological linkage supports the view that photochemical background conditions in tropical marine regions are not only chemically pristine but also dynamically regulated by aerosol–cloud interactions.

To quantify the role of lightning-produced nitrogen oxides (LNOx) in modulating tropospheric ozone over Palau, we performed two sensitivity simulations using the GEOS-Chem model by turning off and doubling LNOx emissions (Fig. ). As shown in Sect. , lightning activity is negligible in the TWP region. Therefore, substantial perturbations to LNOx emissions, near-surface and free-tropospheric NO mixing ratios over the TWP exhibit little response (Fig. c–i). In contrast, pronounced responses to LNOx perturbations are evident over lightning-active regions such as northern Australia and the Maritime Continent.

The resulting ozone anomalies extend eastward into the TWP, indicating that the ozone response over Palau primarily reflects transport from nearby regions, such as northern Australia and the Maritime Continent rather than local photochemical production. As a result, removing or doubling LNOx emissions (Fig. b and c) leads to modest ozone changes over the TWP, with regional differences of approximately ±5 ppb (Fig. a and d).

Moreover, LNOx emissions moderately enhance regional hydroxyl radical (OH) concentrations (Fig. b and e), reflecting their contribution to atmospheric oxidative capacity. In the absence of LNOx, OH levels decrease by approximately 0.02 ppt. Under clean air conditions such as the TWP, OH is mainly produced by ozone photolysis followed by the reaction of excited atomic oxygen with water vapor . Specifically, the dominant pathway is O3+hν→O(1D)+O2, followed by O(1D)+H2O→2OH . While this pathway governs the primary production of OH, its regeneration depends strongly on the availability of NO through the reaction HO2+NO→OH+NO2. In the absence of NO, the lifetime of HO2 increases, yet OH recycling becomes inefficient , leading to a less oxidizing atmosphere and reduced photochemical ozone production efficiency.

In remote tropical regions like Palau, where anthropogenic NOx emissions are minimal and lightning activity is strongly suppressed, lightning represents one of the few potential sources of NOx. Our sensitivity simulations indicate that, under such conditions, the impact of LNOx on local ozone is indirect and transport-driven rather than in situ. Although the overall influence on ozone is modest, LNOx plays a systematic role in regulating the regional oxidative capacity, consistent with previous assessments identifying lightning as a key natural NOx source in remote tropical atmospheres .