O3 is a common tracer in transport studies within the TWP, with typical dynamical conditions being conducive to a tropospheric column of low O3 VMR attributed to local air masses e.g.,. Dominant marine convection leads to low NOx concentrations and subsequently low O3 production rates e.g.,. In this environment the HOx-driven loss of O3 is favored (see Reaction ), resulting in a lifetime of about 5 d for boundary layer O3 e.g.,. The low-NOx environment is fostered by dominating easterly winds in the annual mean and most of the troposphere . Most of the air reaching Palau has crossed the Pacific and has thus been cleaned of O3 precursors from anthropogenic or other continental pollution. Palau has a hot, humid and wet climate all year, resulting in a continuously high convective activity . This persistent high convective activity creates a well-mixed profile of uniformly low O3 VMR throughout the free troposphere, as illustrated in Fig. a. We refer to this as the clean state of the atmosphere or simply “background”.

Layered structures of higher O3 VMR in the mid-troposphere disturbing the uniform low O3 profile are often observed and a signature of non-local air masses e.g.,. Figure b shows an example profile with two distinct layers of enhanced O3 levels (> 50 ppb) around 4 and 8 km, interrupting otherwise quite constant levels of low O3 VMR (∼ 25 ppb). Enhanced tropospheric O3 levels are generated either by photo-chemical processes in high tropospheric NOx regimes, i.e., in polluted air masses, or by photo-dissociation of O2 in the stratosphere and subsequent transport to the troposphere. In situ net O3 production is unlikely at this altitude in the remote TWP, which is far from pollution sources and shows a lack of lightning activity . Free tropospheric O3 lifetimes are on the order of weeks and comparable to the dynamical timescales of long-range transport within the tropics and from the subtropics e.g.,. Thus, air masses transported either from polluted areas elsewhere or the extratropical stratosphere can retain their high O3 VMR until arrival in the deep tropics.

To complement O3 as an indicator of air mass origin, we made use of RH as a measure of vertical displacement of air masses. The benefit of using RH is the readily available data from the combined sonde measurements. The dominant anti-correlation between O3 and RH in the PAO dataset further suggests closer examination of RH (Figs.  and ). RH allows us to distinguish between different underlying processes affecting either local or non-local air masses e.g.,. Local air masses of low O3 VMR often show high RH values (O3-RH+) in accordance with the assumed dominance of convective activity and uplift. The example in Fig. a shows that RH values characteristically decrease with altitude but remain greater than 45 % throughout the mid-troposphere cf.. Non-local, ozone-rich air masses often correspond to low RH compared to the air masses above and below the humid tropical column (O3+RH-). The depressed RH levels can be explained by either stratospheric origin or various dynamical processes during transport, mostly involving a descent of air masses towards Palau that results in adiabatic compression and heating of the air, reducing RH at constant absolute humidity . The study of RH data thus supports our understanding of processes governing O3 variability above Palau.

The competing hypotheses for the genesis and origin of O3+RH- air demand the use of an additional tracer e.g.,. In this study, we use potential vorticity (PV) as a dynamical tracer to identify a potential stratospheric origin of air masses. We assume a significant in-mixing of extratropical stratospheric air for air masses with an absolute PV greater than 1.5 PVU for at least 1 d during 10 d before arrival in Palau .

In conclusion, we propose four qualitative categories of air masses and five different pathways identifiable by our tracers, as illustrated in Fig. . Ozone-depleted air masses (O3-, turquoise colors) are of local or Pacific convective origin, and ozone-rich air masses (O3+, brown colors) originate from non-local pollution or the stratosphere. High RH results from a dominant convective uplift of air masses (RH+, darker hues), while low RH indicates a stratospheric origin or dehydration of previously lifted air masses during transport due to clear-sky cooling and descent (RH-, lighter hues). O3+RH- air is characteristic for two different pathways (solid and dotted light brown lines in Fig. ), thus representing two different types of air masses.

As discussed above, the two important air mass categories are tied to the dominant anti-correlation of O3 and RH above Palau: O3-RH+ background air and O3+RH- air masses occurring mostly in layers interrupting the background. Air masses with positively correlated O3 and RH are observed less often and are more difficult to assess. O3+RH+ air masses (dark brown) are potentially caused by non-local pollution convectively lifted in the source region and transported rapidly towards Palau on the same altitude, conserving the high humidity. We further suggest that O3-RH- air (light turquoise) consists of clean boundary layer air convectively lifted in the Pacific vicinity of Palau, dehydrated due to clear-sky cooling during transport. Other processes, like remoistening or in-mixing of midlatitude air during transit to the TWP could also play a role and are difficult to assess just using the tracer data.

To detect and assess anomalous layers from balloon and aircraft profiles, different methods have been applied . used data from three NASA campaigns (PEM-Tropics, PEM West A and B) to calculate a free tropospheric background mode for various atmospheric constituents (O3, H2O, CO, CH4) and for individual profiles. Using a spike detection approach, they produced extensive statistics for the frequency of anomalous layers in the whole tropical Pacific region for two different seasons. The same methodology was applied to MOZAIC aircraft data by . Its main disadvantage lies within an arbitrary threshold used to determine the background mode. Despite the statistical uncertainty, recognized the importance of anomalous layers in tropical Pacific profiles and emphasized their role in atmospheric chemistry modeling due to their frequent occurrence.

A similar analysis of longer time series from three SHADOZ sites Southern Hemispheric Additional Ozonesondes Network; confirms the frequent occurrence of dry enhanced O3 layers below 12 km . Using a statistical method for the layer detection, the study found respective layers in approximately 50 % of profiles per year and station with differing seasonal variations. The study relies on the separation of a wet and distinct dry season at the respective stations, which is not applicable for the Palau site see.

isolated the low O3 background or “primary mode” by removing all “dry” data with RH less than 45 % for all observations during the CONTRAST campaign . They thus refrained from resolving the individual vertical structure of layers but proposed the RH threshold as an overall criterion in the free troposphere to separate local and non-local air masses. emphasized the possible variability of air mass origin within the individual vertical column, which is obscured in seasonal or even annual profile averages, i.e., layered structures in the O3 profile hidden within the typical so-called “S” shape (cf. Sect. ). Their method, however, seems limited to the specific sites and season of the CONTRAST campaign, as it does not yield a robust output for the Palau data during all seasons .

For a quantitative definition of air masses above Palau, we first determined background profiles in the free troposphere from the PAO O3 and RH time series. These profiles represent humid, ozone-poor, local air masses that are controlled by convective influence and not by long-range transport. This was performed on a statistical and monthly basis and aimed to identify profile shapes similar to the signature profiles in Fig. a. For the monthly O3 VMR background profiles we chose the 20th quantile (Q20) profiles. Q20 guarantees sufficiently low O3 values to avoid a “belly” shape in the profiles, typical for tropical average O3 profiles. For the monthly RH background profiles we chose the 83.3th quantile (Q83) profiles. Q83 is the upper boundary of the central 66.6 % range, guaranteeing high humidity. Figure  shows an example of these background profiles for March. The monthly quantile profiles have been vertically smoothed using first a 1 km binning and then exponentially weighted moving averages with altitude (see also Figs.  and ). Figure  includes the median and central 66.6 % ranges for both tracers separately and is calculated from 16 individual profiles seefor more details.

Our method roughly followed the approach of . However, defined air masses above the 83.3th quantile as O3 enhanced layers. With the Q20 limit for O3 we used a less conservative approach because the Palau background atmosphere is characterized as a uniform, well-mixed low O3 profile, caused by uplift of ozone-poor boundary layer air in active convection. We chose this particular O3 quantile to yield the most uniform, “straight-line” profile in the free troposphere for all monthly averages of the given time series. This essentially smoothes out any interrupting layers occurring at varying altitudes in individual profiles. While analyzed RH in ozone-enhanced anomalous layers in contrast to the air masses above and below, we applied the statistical approach to both tracers and chose Q83 for RH to detect typical convection associated with the background. The resemblance of both the particular low O3 and high RH quantiles with the example of an individual background profile in Fig. a is evident. The monthly averaging accounts for the seasonal variability of the background as the uniform O3 profile shifts towards higher base VMR from summer–fall to winter–spring (cf. Figs.  and ).

In a second step, we determined the anomalies against the monthly background profiles, denoted in the following by ΔO3 and ΔRH. These are calculated for individual measured O3 / RH pairs within each sounding and in their respective month–altitude bin. Within the anomaly space, we define a new background air mass category, “ΔO3∘ΔRH∘”, with close to zero anomalies in both tracers. We further denote eight other quantitative air mass categories with combinations of positive (“+”), negative (“-”) or close to zero (“∘”) anomalies for both ΔO3 and ΔRH individually (Sect. ). In this study, however, we limited further analyses to the two most interesting quantitative air mass categories, ΔO3∘ΔRH∘ and ΔO3+ΔRH-, i.e., per definition humid, ozone-poor background air masses and anomalously dry, ozone-rich air masses. In contrast to we did not assess the vertical structure of the anomalous layers in individual profiles. Hence, neither the layer thickness nor the layer position within the profile were considered. Any measured data point of a profile was quantitatively attributed to one of nine air mass categories, which were then analyzed “in bulk”. In our following analysis, we further focused on the altitude region between 5 and 10 km, where the weakest cloud mass divergence occurs, i.e., the weakest convective detrainment and the greatest frequency of anomalous layers.

Figure  shows time–height cross-sections of the annual variability of tropospheric O3. Figure a shows monthly means, Fig. b shows anomalies from the annual mean profile and Fig. c shows the annual mean profile of the overall time series (solid black line), excluding El Niño 2016. Panel (c) also shows the standard deviation (SD) and the relative standard deviation (RSD) of the overall annual mean profile (dashed and dotted lines, respectively), as well as the annual mean profiles of individual years (colored lines). As the fraction of the SD relative to the annual mean, the RSD can be considered as a measure of variability with altitude cf.. The annual mean tropospheric O3 VMR profile (Fig. c) has an S shape typical for tropical sounding stations e.g.,. O3 VMR are lowest in the boundary layer (< 20 ppb) and also low between 10 and 12 km (< 30 ppb). They increase in the mid-troposphere to about 35 ppb and again above 12 km towards their stratospheric maximum, which is an order of magnitude higher. The interannual variability of the annual mean O3 profile is low for all years, with the exception of 2016.

The monthly O3 VMR (Fig. a and b) reveal two dominant signals: an annual cycle in the 5–10 km altitude range with the maximum from February until April (30–60 ppb) that is 2–3 times greater than the minimum from July until October (10–30 ppb) and a reverse, strong cycle in the TTL with maximum anomalies (> 40 ppb) from June until September. The peak in RSD of 50 % at about 17 km reflects the strong TTL cycle. The enhanced variability in mid-tropospheric O3 (RSD ∼ 30 %) corresponds with the cycle revealed by the monthly means. Annual variations in O3 observations are smallest between 10–12 km and especially in the boundary layer (RSD ∼ 20 %), and these are thus coincidental with the O3 minima of the annual profile. The regionally typical ozone-poor background is clearly subject to seasonal variations.

A more detailed analysis of the seasonal O3 variability helps to identify recurring seasonal characteristics and differences in controlling processes like synoptic or meteorological conditions. Therefore, we divided our time series into four seasons, shifted by 1 month compared to the temperate climate seasons: November–December–January (NDJ), February–March–April (FMA), May–June–July (MJJ) and August–September–October (ASO). These Palau seasons reflect different influences dependent on the time of the year and were chosen empirically by sorting monthly O3 profiles Fig. A1 by similar shape, considering the full free troposphere. The resulting four most different types of profile shapes are shown in Fig. . The Palau seasons turned out to be centered around the equinoxes, which is reasonable for a tropical station.

In the 5–10 km altitude range FMA and ASO show the largest differences and represent the extremes of the mid-tropospheric cycle. The S shape of the annual mean profile prevails in the winter seasons, i.e., in NDJ and most pronounced in FMA. The interannual variability is particularly high during FMA from 5 to 10 km with generally enhanced O3 VMR between 40 and 50 ppb (Fig. ). The ASO profile is closest to the uniform background profile (20–25 ppb O3 VMR) with little interannual variability, while having the highest sampling rate (59 profiles). We thus refer to ASO as the background season. In the TTL, both extreme seasons, ASO and FMA, show a pronounced O3 minimum, except during the 2016 El Niño FMA season (Fig. ). The steep onset of higher levels of (stratospheric) O3 VMR in ASO occurs at lower altitudes compared to the NDJ and FMA seasons.

MJJ and NDJ can be considered as intermediate seasons with respect to the mid-tropospheric cycle; i.e., O3 values here are in between the minimum and maximum. The NDJ profile resembles the annual mean profile up to about 14 km, while the MJJ profile diverges from the annual mean above 10 km towards higher values. The resulting tilted line shape during MJJ, compared to the straight line profile during ASO, implies a more gradual increase of O3 in the upper troposphere–lower stratosphere (UTLS) cf. and a lack of a pronounced O3 minimum in the UT. In MJJ, the tropopause is at its lowest altitude of the year, which is also the case for the occurrence of high levels of stratospheric O3 VMR.

The application of our method to separate air masses on the full PAO time series for the free troposphere (3–14 km) yields the distribution of anomalies from the background profiles, ΔRH and ΔO3, given in Fig.  (see also Sect. ). The 2D histogram in Fig. a shows ΔO3 versus ΔRH from all months and altitudes as percentages of the total count of all measured free-tropospheric O3 / RH data pairs in greyscale. Data along the zero lines within this anomaly space indicate zero anomalies and thus an attribution to the respective background profiles, Q20 of monthly O3 VMR and Q83 of monthly RH values. The marginal 1D histograms show the distributions of both tracers individually normalized to unity.

Both ΔRH and ΔO3 show heavily skewed distributions, with a long tail in the ΔO3 distribution towards less frequent high O3 occurrences and even a bimodal distribution for ΔRH with a secondary maximum in the tail towards dry air masses. These distributions motivate the separation of air masses into three domains for both parameters, with one corresponding to the low tail, one to the bulk of the observations and one to the high tail. Overall this leads to nine quantitative air mass categories in a 3 × 3 grid that is referred to as the quantitative grid (Fig. b). The boundaries were chosen empirically with respect to the shape of the distributions with a range of -5/+15 ppb ΔO3VMR and -20 %/+5 % for ΔRH for the central group.

In particular, the correlated occurrence of air masses with high O3 and low RH (ΔO3+ΔRH-) stands out as a low but separate population in Fig. a (lower-right quadrant), while the central group (ΔO3∘ΔRH∘) represents the most frequent conditions and is referred to as background category (not to be confused with the background profile) since anomalies from the background profiles are smallest. In this study, we focus on the distinct population of ΔO3+ΔRH- air masses in contrast to this background category. ΔO3∘ΔRH∘ air masses are characterized by significantly higher RH and lower O3 and are therefore also referred to simply as the humid, ozone-poor background.

The anomalies are mostly clustered within 20 ppb along the ΔO3 zero line indicating an overall dominance of ozone-poor air masses. Apart from the tail towards less frequent higher ΔO3, the narrow distribution is nearly Gaussian and centered around 0 ppb ΔO3 VMR with a tendency towards positive values. The majority of ΔRH values appear between -60 % and 20 %. The bimodality of the ΔRH distribution is persistent in all seasons with the exception of MJJ (Fig. ). The primary, dominant mode is centered slightly off the ΔRH zero line towards negative values, i.e., the majority of measured RH values match the Q83 or slightly lower values of the respective month–altitude bin. The secondary mode occurs for ΔRH values 40 % lower than their background profile estimates.

Figure  shows the ΔO3 / ΔRH distribution for the seasons FMA and ASO. As expected from the previous analysis, ΔO3+ΔRH- are almost absent in ASO (Fig. a), emphasizing that the season represents an overall ozone-poor, mostly humid background in the free troposphere. In contrast to the seasonal mean FMA O3 profile with elevated O3 levels in the mid-troposphere (Fig. ), the FMA anomalies distribution (Fig. b) reveals a dominance of background category air masses over ΔO3+ΔRH- air masses within the free troposphere. A statistical view on the seasonal occurrence of different air masses in the mid-troposphere emphasizes the year-round dominance of ΔO3∘ΔRH∘ air, which were observed in more than 70 % of profiles within each season (Fig. ).

In the following, we focus on the 5–10 km altitude range with the most frequent occurrence of ΔO3+ΔRH- air masses and the largest seasonal differences in O3 VMR (cf. Fig.  and Sect. ). A footprint of air mass transport to Palau is analyzed using 10 d backward trajectories for the study period (2016–2019) sorted by season arriving in Palau in the 5–10 km altitude range (Fig. ). For better visualization, a representative subset was chosen, displaying only every 20th trajectory of a profile (roughly one trajectory per kilometer). Colored line segments show three different time periods backwards, 3, 5 and 10 d, thus indicating the velocity of air mass transport.

Most air masses reach Palau from the east but have traveled two different paths on different timescales. The seasonal distinction of trajectories roughly separates an eastern Pacific pathway, dominating in MJJ and ASO, from an anticyclonic route in NDJ and FMA, connecting Palau with Southeast Asia and some remote areas as far as the African continent. The anticyclonic route can be associated with long-range transport on shorter timescales compared to the Pacific route. As these transport patterns are most pronounced in FMA and ASO, respectively, which also exhibit the greatest tracer differences in the mid-troposphere (cf. Fig. ), we focus on these two Palau seasons in the following.

The geospatial extent of the footprints differs significantly for season FMA and ASO. While all trajectories stay mostly within the tropical zone between 0 and 30° N, trajectories during FMA have a wider longitudinal extent than during ASO. Most ASO air masses never left the Pacific Ocean area within 10 d before arrival in Palau and have traveled shorter distances. Only a few trajectories on a Southern Hemispheric cyclonic route show the influence of the western Pacific monsoon, which is active from July to October but governs air mass transport mostly below 5 km altitude .

Figure  shows the location of air masses 5 d before their measurement in Palau inferred from trajectories for the seasons FMA (top) and ASO (bottom) as an indication of the origin of the air masses. The geospatial probability density function of data points is highlighted as black contour lines in Fig. a. Here, the colors indicate O3 VMR as observed upon arrival in Palau. Figure b and c show the same locations as Fig. a but are color coded by RH (Fig. b) or difference in pressure altitude (Fig. c). Red colors in panel (c) indicate descent, while blue colors indicate ascent towards Palau.

The dominant source region of free-tropospheric Palauan air masses, indicated by the center of the density distribution in Fig.  a, is located east of Palau during both seasons and corresponds to low O3 VMR (≤ 30 ppb). During FMA, this eastern center of the distribution is actually split in two parts along a latitudinal axis. The dominant part is located further south on the Equator, exhibiting low O3 VMR. The northern part of the eastern center is of higher O3 VMR. A separate cluster of trajectory points showing enhanced O3 VMR (≥ 60 ppb) exists northwest of Palau during FMA, roughly extending from India to Taiwan. As revealed by the full 10 d trajectories in Fig. b, all air masses of higher O3 VMR during FMA can be related to the anticyclonic route, though some took longer than 5 d to travel from Southeast Asia to Palau. During ASO the overall unimodal distribution is centered on Palau's latitude, 20° to the east, and observations of enhanced O3 VMR originate outside the main cluster. For both seasons, air masses with O3 VMR greater than 60 ppb rarely originate south of Palau.

The distinction between the two pathways is also reflected in RH (Fig. b) and even more clearly in the vertical displacement of air masses from origin to destination (Fig. c). In FMA, Fig. c shows descended air masses originating north and ascended air south of Palau. Here, the descent of air masses corresponds with greater O3 VMR and lower RH, although the pairwise correlation, R, of these two parameters with the pressure height difference is not strong (R=-0.53 for O3VMR and R=0.44 for RH, for the latter cf. Fig. ). We can, however, associate these dry, descending air masses with the anticyclonic route shown in the 10 d backward trajectory footprint.

For ASO air masses, the picture is not as clear, but ascent and humid air masses with low O3 VMR are slightly dominating (Figs. c and b) and have likely followed the Pacific pathway. The pairwise correlation for pressure–height difference and O3 VMR or RH, respectively, is even lower than during FMA (R=-0.36 for O3, R=0.35 for RH). However, there are clearly two main groups of air masses, one with RH below 40 %, descended mostly by only 1 km, and one group with RH centered around 70 %, ascended mostly by 1 km or more.

The investigation of the history of PV for Palauan air masses revealed no stratospheric pathway. Below 1 % of all 13 627 trajectories stayed above 1.5 PVU for more than a day during 10 d transit to Palau. Within 5 d before arrival only a few trajectories crossed the 1.5 PVU threshold.

While we have looked at the origin of the trajectories as a function of season in the previous section, we now turn to the origin as a function of air mass. The classification of trajectories by quantitative air mass category as defined in Sect.  yields the distribution of source regions as shown in Fig. . Background ΔO3∘ΔRH∘ air masses originate within a compact region in the Pacific, mostly east of Palau and stretching as far as 180° E. The origin pattern for ΔO3+ΔRH- air masses clearly shows transport from the Southeast Asian region and along the anticyclonic path to Palau. In comparison to the seasonal analysis, the classification of trajectories by air masses shows a clear distinction for the history of vertical displacement of the air masses. On their way towards Palau most ΔO3∘ΔRH∘ air masses have been ascending, while ΔO3+ΔRH- air has descended.

A seasonal analysis of the trajectories sorted by air mass anomalies reveals some differences in the source regions of ΔO3+ΔRH- air masses (Fig. ). The origin hotspots around India and the Southeast Asian peninsula are almost exclusively observed during FMA. The southern Philippines and the area slightly northeast of Palau are equally dominant source regions in this season. In ASO, ΔO3+ΔRH- air masses are less frequently observed, and their origins are clustered around three main areas: (1) northeast of Palau in the western Pacific, (2) between Borneo and the southern Philippines, and (3) in the New Guinea region, extending towards northern Australia.