Figure 7 illustrates ozone climatology obtained by compilation of 15 years of data recorded between January 1998 and December 2012. Three regions are explored: (a) equatorial, (b) tropical and (c) subtropical. Over these selected regions, total ozone exhibited high values from July to November. A positive gradient was observed from May to the maximum annual peak in September (equatorial region) but occurred a month later in October for tropical and near-subtropical regions. The recorded positive gradient over the subtropics was the largest and this indicated a high seasonal variability in this region compared to tropical and equatorial zones. Low ozone levels were recorded during the austral summer/autumn period over the three regions. Ozone recorded in the equatorial region was high in March, April and May compared to the tropics. Two peaks were clearly observed for the equatorial region, and these highlight a semi-annual cycle with a maximum in September–October and a second maximum in March–May.

These results can be explained as follows: stratospheric ozone is formed through a photochemical reaction (Chapman, 1930) that requires a sufficient quantity of solar radiation. The equatorial region is therefore the primary site of ozone production (Coe and Webb, 2003). During the period close to equinox, solar radiation leads to an increase in ozone production over the equatorial region. During summer, ozone transport consists primarily of vertical upwelling movements which are confined to the low-latitude region. This results in less ozone being distributed to the subtropics. This effect is the main contributing factor to why a total ozone peak is observed in autumn over the equatorial region (Mzé et al., 2010). However, the semi-annual equinox process may not be the sole explanation for the maximum value recorded in the equatorial region during September spring equinox, because it is generalized over the three regions. High ozone levels recorded in winter/spring are due to transport and accumulation of ozone in the Southern Hemisphere as a result of Brewer–Dobson circulation on a regional (air mass transport from the tropics to subtropics) and global scale (from summer hemisphere to winter hemisphere) (Holton et al., 1995; Fioletov, 2008). Transport of ozone from the tropical to extratropical regions is the most dominant process during the winter period (Portafaix, 2003) and constitutes the principal reason for the observed annual maximum over the region 20–30∘ S. The decline of ozone levels during late spring and summer is explained by a decrease in ozone transport coupled with ozone photochemical loss (Fioletov, 2008). However, the observation of maximum ozone in September over the equatorial region but 1 month later over the subtropics may be explained by a delay of ozone transport between the two regions. As shown in Fig. 7, the curve of the seasonal distribution of TCO attributed to (b) the tropical region is below that of (a) the equatorial zone from January to September. While considering the process of formation and transport of ozone as described above, TCO annual records in tropical regions should be higher than that recorded over equatorial regions. However, it is important to note that both sites (Samoa and Fiji, in the 10–20∘ latitude range) are located in the Pacific Ocean. The low TCO quantity observed in the Pacific is due to a low tropospheric ozone levels over the region as a result of a zonal Wave-One observed on tropical tropospheric ozone (Thompson et al., 2003a and 2003b). Similar results were reported by Thompson et al. ( 2003a and 2003b) where low ozone levels over SHADOZ Pacific sites were observed with respect to African and Atlantic sites due to the zonal Wave-One.

The contribution of seasonal cycles (annual and semi-annual oscillations) to ozone variability obtained from the regression analysis is presented in the first and second columns of Table 4. The calculated values show that except for Nairobi, the annual cycle constitutes the predominant mode of ozone variability over the studied sites and its influence is strongly observed over the subtropics. The contribution of annual oscillations to ozone variability exhibits a latitudinal signature, with its minimum and maximum at equatorial and subtropical zones respectively. The percentage contribution of annual oscillations to TCO variability decreases equatorward from Irene (65.96±3.93 %) to Nairobi (20.33±1.89 %). As explained in Sect. , the annual cycle is modeled by a sinusoidal function with maximum and minimum in winter/spring and autumn/summer. This kind of annual variation mode characterizes the ozone climatology of southern tropics and subtropics (Toihir et al., 2013, 2015a). The response of annual oscillations to ozone variability obtained for individual sites and reported in Table 5 is positive. This indicates that the modeled annual cycle function is in phase with the original ozone data. However, the amplitude of the annual cycle on total variance of ozone over the subtropics is the highest, and this indicates that ozone variance is sensitive to annual oscillation in the subtropics compared to tropics. The amplitude was evaluated at approximately 5.6 % by unit of the standardized annual function in subtropical sites (Reunion and Irene) and varied between 1 and 3 % from Fiji towards the Equator.

Results presented in Table 4 show that on average, 13.86±1.26 % of ozone variability can be explained by semi-annual oscillations over stations located at 5–10∘ S. However, the semi-annual pattern was not apparent over Nairobi compared to the rest of the equatorial sites. The predominant mode of ozone variability in this equatorial site was the QBO (20.48±1.92 %), followed by the annual oscillations (20.33±1.89 %). The semi-annual oscillation pattern of TCO was observed to be significant at low latitude, with its maximum contribution over Ascension Island. The response values associated with semi-annual oscillation to ozone variability are presented in column 2 of Table 5 and show that a maximum in amplitude over Ascension Island was observed. Both quantities (contribution and response) decreased gradually while moving away from the Ascension Island site towards the Equator or southward. This indicates that the semi-annual oscillations of ozone are weighted at approximately 8∘ of latitude and coincide with the location where the influence of equinox processes on ozone variability is the greatest.

The QBO constitutes the second most important parameter influencing ozone variability after seasonal oscillations. The QBO behaviour on ozone time series can be observed by removing the monthly climatological value from its monthly mean. In this investigation, the deseasonalized ozone signal was further subjected to 3-month smoothing in order to filter out smaller perturbations from intra-seasonal oscillations and to address the component due to ozone biennial oscillations.

Figure 8 illustrates the deseasonalized ozone (blue dashed curve) time series (1998–2012) for equatorial and subtropical regions. These data are superimposed with the monthly mean zonal wind data at 30 hPa over Singapore (black curve). From Fig. 8 it can be seen that the predominant variability of deseasonalized monthly ozone data is an approximately 2-year cycle linked to the QBO and represented by the zonal wind data at 30 hPa. However, the equatorial and subtropical ozone signals seem to be opposite in phase. This indicates that if an excess of ozone is recorded over the equatorial region, the subtropical zone shows a trend illustrating a deficit with respect to the monthly climatological value and vice versa. The QBO change from the Equator to the subtropics is due to a secondary circulation having an upwelling branch in the subtropics and a subsidence branch in the tropics during the westerly phase of the QBO (Chehade et al., 2014). This secondary circulation is characterized by a deceleration or an acceleration of the Brewer–Dobson circulation (BDC) during the westerly or easterly phase of the QBO respectively. The deceleration of the BDC leads to an increase in ozone in the tropics, while the acceleration leads to a decrease in ozone in the tropics. Due to this circulation, the deseasonalized TCO data over the equatorial region are in phase with the QBO index. Positive ozone anomalies correspond to the QBO westerly phase and negative anomalies are linked to the QBO easterly phase. Similar results have been determined in previous studies (Randel and Wu, 2007; Butchart et al., 2003; Zou et al., 2000; Bourassa et al., 2014; Peres et al., 2017).

The above process was explained by Peres et al. (2017) and may be outlined as follows: the descending westerly phases of the QBO are associated with a vertical circulation characterized by downward motion in the tropics and upward motion in the subtropics. This leads to a weakening of the normal speed of the Brewer–Dobson circulation. In this way the upward motion of air mass is slowed down and the tropopause height decreases. As the ozone mixing ratio increases with the increase in altitude in the lower stratosphere, ozone production can occur for a longer period than normal (Cordero et al., 2012). This mechanism leads to a positive column ozone anomaly at low latitudes and a negative anomaly in the subtropics. During the QBO descending easterly phase, ozone formed spends less time at low latitude due to the enhanced Brewer–Dobson circulation (Cordero et al., 2012). The newly created ozone is rapidly transported to the subtropics, resulting in a negative ozone anomaly at lower latitudes and a positive ozone anomaly in the extratropics.

Butchart et al. (2003) explained that the equatorial ozone anomalies due to QBO forcing extend to 15–20∘ of latitude. In order to investigate the border between equatorial and subtropical QBO signals on ozone, the response of the QBO signal to TCO variability for individual sites was calculated based on the regression coefficient associated with the QBO on the Trend-Run model (Eq. 4). The response values are given in column 5 of Table 5. It is seen that the response of QBO to TCO variability by unit of normalized QBO signal is positive when the QBO is in phase with the ozone signal and negative for the reverse case. In this study a positive response (of decreasing magnitude) is obtained with the decrease in latitude moving from Samoa (14.13∘ S) to Nairobi (1.27∘ S). Conversely the responses obtained are negative at Fiji (18.13∘ S), Reunion (21.06∘ S) and Irene (25.90∘ S). These results indicate that the border between the equatorial and subtropical QBO-ozone signature is found at around 15∘ in latitude. Similar results were reported by Chehade et al. (2014) by calculating the coefficients of regression associated with the influence of the QBO on ozone variability. The Chehade et al. (2014) study reported positive and negative regression coefficients between 0∘ and 15∘ S and from 15∘ S towards the pole respectively.

In this work, positive ozone anomalies with amplitude varying between 7.5 and 13.9 DU were recorded over the equatorial sites during the zonal wind westerly phase, while the negative anomalies observed during the easterly phase oscillated between -13.3 and -2.80 DU. These results indicate that inter-annual oscillations of TCO over the equatorial regions have maximum amplitude varying within ±14 DU during the study period (1998–2012). This is in agreement with the results of Butchart et al. (2003). For 1980 to 2000 they obtained a deseasonalized ozone signal which varied within ±14 DU over the Equator.

It is worth noting that the maximum amplitude of ozone anomalies observed in the subtropics is lower (within ±11 DU) than the equatorial region (±14 DU). These amplitude values indicate that the QBO modulation on inter-annual variability of ozone is higher over the equatorial region in comparison to the extratropics. Values reported in Table 4 show a clearly decreasing QBO contribution poleward, i.e. from Nairobi (20.48±1.92%) in the equatorial zone to Irene (3.60±1.07%) in the subtropics.

Column 6 of Table 4 presents the percentage of ENSO contributions to inter-annual ozone variability for individual sites. It is clear that the ENSO parameter does not exert much influence on ozone variability over the subtropical sites (Reunion and Irene), where its contribution to TCO variability is assessed to be less than 1 %. However, the ENSO contribution is more pronounced over tropical and low-latitude sites located between 19∘ S and the Equator. In this latitude band, the ENSO contributions observed at Fiji, Samoa, Ascension Island and Java are high compared to Natal and Nairobi. An average ENSO contribution of 9.40±0.65% to total ozone variability over Fiji, Samoa, Ascension Island and Java is observed, while a corresponding value of 5.73±0.75% for Natal and Nairobi is seen (see Table 4). These results indicate that the ENSO influence on total ozone variability is high over the tropics compared to that in the equatorial zone and subtropics. It has been reported by Randel et al. (2009) that ENSO originates in the tropics and is linked to coupled atmosphere–ocean dynamics. This factor may be the main reason for the observed high ENSO contribution over sites located in the tropics compared to low-latitude and subtropical sites. Zerefos et al. (1992) removed the QBO, seasonal and solar cycles from the ozone signal at Samoa and obtained a good correlation between ENSO and ozone signal. They mentioned a possible ENSO–ozone relationship beyond the tropical region only during very large ENSO events (e.g. 1982–1983, 1997). In this study, the low estimated ENSO contribution (less than 1 %) for the subtropical region may also be explained by the fact that there were few large ENSO events recorded during the investigated period (1998–2012).

The tropical sites located in the Pacific Ocean (Samoa, Fiji and Java) were more influenced by ENSO when compared to tropical and equatorial sites (Ascension Island, Natal and Nairobi) due to the high ENSO activity usually observed in the Pacific Ocean (Randel and Thompson, 2011; Chandra et al., 1998, 2007; Logan et al., 2008; Ziemke and Chandra, 2003; Ziemke et al., 2010). In addition, the Ascension Island ENSO contribution was higher compared to those observed at Natal and Nairobi. As ENSO is a coupling ocean–atmosphere event generated in the Pacific Ocean (Ziemke and Chandra, 2003; Rieder et al., 2013), the low ENSO contribution recorded over Natal and Nairobi compared to Ascension Island can be explained by their continental location. As Ascension Island is located in the Atlantic Ocean, the high ENSO contribution observed at this site is probably due to a Pacific–Atlantic Ocean connection.

The ENSO response to ozone variance is assessed and shown in the last column of Table 5. The aim of this work is to define the mean behaviour of ozone variance for individual sites during ENSO events. It is worth noting that the mean ENSO influence is observed on tropospheric ozone (Randel and Thompson, 2011; Ziemke and Chandra, 2003). However, due to weak variability in stratospheric ozone over the tropics (Ziemke et al., 1998) during periods of normal conditions, tropospheric ozone changes were shown to affect the stratospheric ozone and total column ozone variability over the tropics during ENSO events (Rieder et al., 2013). As the MEI is characterized by positive values during El Niño, a positive ENSO response indicates an increase in total column ozone (TCO) and a negative response indicates a decrease in TCO. The ENSO response to ozone variability over subtropical sites (Reunion and Irene) is positive, while it is generally negative over the tropics (except over Java and Nairobi). These results are in good agreement with those reported by Chehade et al. (2014) which are obtained using a regression analysis model. Through their study they obtained negative (positive) ENSO regression coefficients indicating a negative (positive) ENSO response to ozone variability over the tropics (extratropics). It has been mentioned in several papers (Rieder et al., 2013; Randel et al. 2009; Randel and Thompson, 2011; Frossard et al., 2013; Chehade et al., 2014) that a negative ENSO response to ozone variability in the tropics is linked to enhanced transport of an ozone-rich air mass from the tropics to the extratropics due to strengthening of the Brewer–Dobson circulation in the stratosphere during ENSO warm events. As reported by Rieder et al. (2013), ENSO events are associated with more frequent stratospheric warming, an increase in tropopause height and a decrease in stratospheric ozone in the tropics. By comparison, in the subtropics, the tropopause height decreases and stratospheric ozone increases. This effect could be the reason for the observed ENSO positive response over Irene and Reunion (subtropics). The ENSO warm events produce suppressed convective movements over the western Pacific, leading to a positive anomaly of ozone in this region, while the air mass convection is enhanced over the eastern Pacific that leads to a reduction of ozone (Ziemke and Chandra, 2003). For the period 1970–2001, Ziemke and Chandra (2003) obtained a positive El Niño response corresponding to an average peak of positive ozone anomaly evaluated at 5 DU over the Indonesia area. Furthermore, they found a negative ozone anomaly in the eastern Pacific, the location of Fiji and Samoa. According to Ziemke and Chandra, (2003), this could be the main reason for the observed positive ENSO response at Java in the western Pacific and negative response over Samoa and Fiji in the eastern Pacific. The connection between the western Pacific and the Indian Ocean could be among the reasons for the observed positive ENSO response to ozone variability at Nairobi in the eastern African and Reunion in the western Indian Ocean region.

Variation in total ozone concentration also depends on solar flux intensity over the tropics and subtropics. Figure 9 illustrates the variation in annual total ozone anomaly calculated between 1998 and 2012 over the three regions (bottom panel) and the annual 10.7 cm solar radio flux variation during the same period (top panel). Inspection of Fig. 9 illustrates the effect of the 11-year solar cycle on inter-annual total ozone variations. Negative ozone anomalies are observed during periods of low sunspot intensity, thereby confirming the dependence of ozone production on incident solar radiation flux. The converse is also true; that is, high annual total ozone levels are associated with periods of high sunspot intensity. The results presented here are in good agreement with Soukharev and Hood (2006). The latter obtained negative (positive) monthly mean ozone anomalies during periods with low (high) Mg II solar UV intensity. The dependence of annual ozone levels on annual solar radiation is also highlighted by Efstathiou and Varotsos (2013). In this investigation, the high levels of TCO were observed over the tropics when solar activity was at a maximum (2001) and compared to values recorded during a minimum in solar activity in 2008. This illustrated the impact of the solar cycle on global ozone concentrations. In the work presented here, considering the year with solar maximum (2001) to the year with minimum solar intensity, a variation in ozone levels was observed corresponding to approximately 1.05, 1.52, and 1.60 % for the (a) equatorial, (b) tropical and (c) extratropical regions respectively. It is expected that the observed percentage change in ozone levels is directly related to solar flux activity. A study by Zerefos et al. (1997) confirmed that the solar flux component in TCO between 1964 and 1994 was approximately 1–2 % over decadal timescales. It can therefore be concluded from the results presented here, together with those of Zerefos et al. (1997), that the solar flux contribution to ozone variability did not vary significantly during the past 6 decades.

As the solar flux index is one of the input parameters in the regression model, the contribution and response of the 11-year solar cycle to total ozone variability were assessed. The effect of changes in solar flux on ozone levels is presented in the fourth column of Table 5. The obtained responses are always positive, indicating that the solar flux index is in phase with total ozone time evolution as reported in the WMO (2014). The increase in ozone occurs when the sunspot intensity increases and vice versa. The average magnitude of the solar flux response to ozone variability is about 1.57 % by unit of solar flux index. The contribution of solar flux to total ozone variability over the study region varies by between approximately 4.38 and 7.81 %. The contribution of the solar cycle is high compared to the 3 % reported by the WMO (2010, 2014) on the global scale. This is probably due to (1) the length of the data records (less than two solar cycles), (2) the time period under investigation (1998–2012) and (3) the studied region (from equatorial to around 40∘ S). However, the contribution values obtained exhibit a minimum at Reunion in the subtropics and a maximum at Ascension Island in the lower-latitude region. These results indicate a high solar flux influence over the low-latitude region in comparison with the subtropics. The contribution of solar flux to total ozone variability decreases gradually by moving away from Ascension Island equatorward or southward. This solar flux contribution pattern is similar to that observed in semi-annual oscillations (see Sect. 4.2.1). From the results presented in this study it can be inferred that semi-annual variations are modulated by solar activity over the tropics and subtropics and that both semi-annual and 11-year solar forcings control ozone distributions.

The contributions of annual and semi-annual oscillations to ozone variability (15 to 30 km) are shown in Fig. 10. Figure 10 reveals that seasonal cycles are the most dominant forcings to ozone variability in the lower stratosphere. The contribution of annual oscillation (AO) to ozone variability is highlighted over the three studied regions and accounts for more than 40 % in the UT–LS (upper troposphere–lower stratosphere). AO maximum amplitudes in equatorial and tropical regions are found around the tropical tropopause layer (TTL) below 20 km and around 22.5 km in the subtropics. These results are in good agreement with recent findings. Gebhardt et al. (2014) showed that in the lower stratosphere (21 km), ozone variability is dominated by AO in the tropics (20∘ N–20∘ S). Furthermore, Eckert et al. (2014) explained that vertical motion in the TTL had a significant impact on ozone variance and that seasonal variations contributed more than 50 % to ozone variability in this layer. A strong annual cycle signature in the UT–LS has also been reported by Bègue et al. (2010). The important contribution of the annual cycle in the UT–LS may be due to the STE (stratosphere–troposphere exchange) processes which occur seasonally in the tropics. These exchanges are linked to the upwelling air mass through the tropopause in summer and the submersion of stratospheric air mass to the troposphere during winter. These air mass exchanges affect the seasonal budget of ozone and temperature from equatorial to middle latitude. As seen in Fig. 10, annual cycle contributions higher than 40 % are found from 15 to 20 km over the equatorial region. However, the annual cycle contributions higher than 40 % observed over the tropical region are found from 17.5 to 22 km. In addition, the maximum annual cycle contribution for the equatorial region is observed at 18.5 km, whilst in the tropics it occurs at 19.5 km. These observations support the existence of upward propagation of the AOs of ozone below 25 km and the amplitude of these AOs increases with the increase in latitude poleward. This may explain the high annual cycle contribution observed in the 15–24.7 km altitude range in the subtropics. The presence of an annual signature in ozone temporal evolution in this wide altitude band (15–24.7 km) in the subtropics is inconsistent with the strong annual variability observed in TCO, as discussed in the previous subsection.

It is worth noting that the seasonal cycles of ozone observed at altitudes below 25 km are linked to dynamical processes, while above an altitude of 27 km, seasonal cycles are modulated by ozone photochemical processes. Contributions of the annual forcing of greater than 10 % are observed at altitudes above 27 km. Here the contribution part of annual oscillation increases with altitude, especially over the tropical and subtropical regions. These results support the existence of an annual cycle signature on ozone temporal evolution over altitudes above 30 km in the tropical and subtropical regions. Eckert et al. (2014) found that ozone variation due to AO was greater than 10 % above 27 km in the 10–30∘ S latitude band. However, this amplitude decreases above 35 km, where ozone variability is strongly controlled by semi-annual oscillation (SAO). Furthermore, the maximum amplitude of the ozone SAO signal observed by Eckert et al. (2014) was centered slightly above 30 km over the tropics. Mze et al. (2010) observed a strong semi-annual cycle in equatorial ozone climatology between 27 and 36 km, with a maximum peak at 31 km, and Gebhardt et al. (2014) observed ozone variability dominated by SAO in the middle (35 km) and upper tropical stratosphere (44 km) in the tropics (20∘ N–20∘ S).

The (SAO) contribution to ozone variability is highlighted in the present work for altitudes higher than 27 km for the equatorial and tropical regions. Above 27 km, the SAO contribution accounts for more than 10 % of total ozone variability over the equatorial region, while this effect is reduced to less than 10 % for the tropics. This confirms the high contribution of SAO to ozone variability at low latitudes with respect to other regions as discussed in Sect. 4.1.1. As supported by Eckert et al. (2014), beyond the tropics the SAO amplitude decreases gradually to near zero. As reported in previous studies, the SAO is linked to change in the zonal wind regime at the equatorial stratopause, where the maximum component appears during solstice (when easterly) and equinox (when westerly) (Belmont, 1975; Hirota, 1978; Delisi and Dunkerton, 1988). This mechanism leads to temperature anomalies corresponding to 2–4 K in the upper stratosphere (Nastrom and Belmont, 1975), thereby driving a change in the rate of ozone production and loss (Maeda, 1984) during the solstice and equinox periods. As demonstrated in Sect. 4.2.1, the equinox period is characterized by maximum peaks of TCO production over the equatorial region, while the solstice corresponds to minimum ozone levels (or ozone loss) over equatorial regions.

QBO contributions to ozone profile variability between 15 and 30 km are presented in Fig. 11 for the three regions, namely (a1) equatorial, (b1) tropical and (c1) subtropical. A strong contribution of the QBO to ozone variability is observed over the 20–28 km altitude range for the equatorial region. Here the QBO is the most dominant mode and accounts for more than 30 % of ozone variability. These results are in agreement with previous work in which a strong signature of the QBO on inter-annual variability of ozone with large amplitudes over the tropics at altitudes of approximately 20–27 km has been reported (Randel and Thompson, 2011; Gebhardt et al., 2014; Eckert et al., 2014). Figure 11a1 shows two strong peaks at 60 % corresponding to altitudes of 21.5 and 25 km. The QBO contribution is important and does not change much over the altitude range 24 to 26.5 km, indicating that the maximum amplitude of the QBO signature occurs in this altitude range, where the degree of QBO influence on ozone variability is approximately invariant. Fadnavis and Beig (2009) found the QBO maximum over tropics at approximately 26 km, while Eckert et al. (2014) found this maximum at 25 km. The QBO signal on ozone variability is linked to the downward propagation of zonal wind throughout the time. This process leads to a positive (negative) ozone anomaly when in the westerly (easterly) mode with a cycle of around 2 years (Randel and Thompson, 2011; Lee et al., 2010; Butchart et al., 2003). Note that the maximum contributions of the QBO over the tropical and subtropical regions (Fig. 11b1 and c1) are less than 12 %. These results confirm the latitudinal signature of the QBO which is expressed by a decrease in its effect on ozone variability poleward. The inter-annual ozone anomaly linked to the QBO time evolution can be positive or negative depending on the altitude range for a given period due to the downward propagation pattern of the QBO with time (Randel and Thompson, 2011). The responses of the QBO index used are presented in Fig. 11 as (a2) equatorial, (b2) tropical and (c2) subtropical regions. Over the equatorial region, the QBO index profile exhibits positive values below 23 km and negative values above. These results indicate a strong contribution of the QBO westerly regime to ozone variability for equatorial regions below 23 km during the studied period (1998–2012). In contrast, the easterly regime had more influence on variability of ozone above altitudes of 23 km. The opposite case is observed in the subtropics, where the QBO index profile exhibits positive values for altitudes between 18 and 23 km and negative values between altitudes of 23 to 27.5 km. Similar results have been reported by Bourassa et al. (2014). The response amplitude obtained in this work over the subtropics varies within ±5 %, whilst it is at ±7 % (by unit of the normalized QBO index) over the equatorial region. These results reaffirm the key role of the QBO in ozone variability highlighted in equatorial regions in contrast to the subtropics.

ENSO contributions to ozone profile variability over the 15–30 km altitude range are presented in Fig. 12 for the three regions (a1) equatorial, (b1) tropical and (c1) subtropical. These calculations show that ENSO events weakly affect subtropical ozone levels. In contrast, the ENSO contribution and its associated response to ozone variability appear to be important below an altitude of 25 km at tropical latitudes. These results are in good agreement with the study by Sioris et al. (2014) in which a statistically significant ENSO contribution was reported at an altitude of 18.5–24.5 km in the tropics. Such results indicate that ENSO influence on ozone variability is essentially focused on the troposphere and lower stratosphere. ENSO contribution to the lower stratospheric ozone is highlighted by the 20–25 km altitude band with peaks toward 15 and 13 % over the tropical (at 23 km) and equatorial (at 21 km) regions respectively. The response linked to this contribution over the equatorial region oscillates with amplitude, varying between -8 and 4 % (Fig. 12, curve a2). Here the ENSO response values are negative below an altitude of 23 km and positive in the 23–26 km altitude range, indicating that the multivariate ENSO index is in phase with ozone time evolution above 23 km and in opposite phase below 23 km. As reported in previous studies (Ziemke and Chandra, 2003; Rieder et al., 2013), ENSO events contribute to the variation in tropopause height associated with decrease/increase in TCO due to the enhancing/suppressing of tropical convection. The results obtained in the equatorial region indicate that the ENSO effect, associated with enhanced convection, generally affects the lower part of the stratosphere. This leads to a decrease in ozone levels below 23 km, while the suppressed convection present generally results in a positive response to ozone variance above 23 km.

However, the amplitude obtained for the negative ENSO response is the largest. These results suggest that ozone variability linked to ENSO events is more sensitive to the enhanced equatorial convection in the UT–LS. In a recent study, similar results were reported by Randel and Thompson (2011), in which they observed a large negative change in ozone variance due to enhanced Brewer–Dobson circulation over the tropics associated with ENSO variability. In addition, the obtained ENSO response over the tropical region (Fig. 12b2) is negative, indicating a decrease in ozone due to ENSO events occurring during the study period. From the troposphere to approximately 20 km, the ENSO signal exhibits an important contribution and response to ozone variability over the tropics as reported in previous published works (Bourassa et al., 2014; Sioris et al., 2014; Randel and Thompson, 2011). However, the response obtained has a larger amplitude (maximum of -38 % at 16.5 km) than those reported by the above-mentioned papers due to the location of the Fiji and Samoa stations (the two sites that represent the tropical region in this work). As mentioned above, Fiji and Samoa are located in the eastern Pacific, a region where ENSO events are significant. Here the ENSO contribution to total ozone variability is observed to be greater than 10 % below 18.7 km. The negative ENSO response associated with this contribution is explained by an enhanced tropical upwelling during ENSO events which led to a decrease in ozone levels in the lower stratosphere of the eastern Pacific Ocean region as reported by previous studies (Randel and Thompson, 2011; Randel et al., 2009; Calvo et al., 2010).

The trend analysis of TCO was performed for each site and results presented in Fig. 13. The TCO trends are assessed based on the Trend-Run model and expressed in percentage per decade. The obtained trend values are positive, suggesting an increase in TCO during the study period. However, the mean values of TCO trends obtained from the Samoa site (14.13∘ S) equatorward are lower than 1 %, while they are higher than 1.5 % for Fiji (18.13∘ S), Reunion (21.06∘ S) and Irene (25.90∘ S). These results illustrate that the increase in TCO in the subtropics is greater than in the tropics. Figure 14 summarizes the trend evolution over the three latitudinal bands. An increase in the trend with latitude southward is observed. The average trends (in percentage per decade) are +0.74±0.6, +1.55±0.4 and +1.74±0.40 (±1σ) over the equatorial, tropical and subtropical regions respectively. These TCO trend results are consistent with the WMO (2014), where a positive trend of approximately 1±1.7 (±2σ) was reported at 60∘ N–60∘ S for the period 2000–2013. These positive trends are probably linked to the decline of effective equivalent stratospheric chlorine (EESC) over the globe as supported by previous studies (Yang et al., 2006; Anderson et al., 2000; Waugh et al., 2001; WMO 2010). However, compared with the tropics and subtropics, the delay in ozone improvement observed in the equatorial region is probably due to the tropical strengthening of the Brewer–Dobson circulation (WMO 2014; Randel and Thompson, 2011).

It should be noted that approximately 10 % of ozone is tropospheric and is produced as a result of chemical reactions between species such as nitrogen oxide (NOx), carbon monoxide (CO) or volatile organic compounds (VOCs). In the context of climate change due to anthropogenic pollutant emissions and in spite of the efforts by the international community to reduce GHG emissions, pollution has continued to increase in the Southern Hemisphere, leading to a systematic increase in tropospheric ozone. Thompson et al. (2014) assumed that growth in ozone precursors like NOx and VOC may account for the characteristic free tropospheric ozone increase, especially in wintertime. The observed positive trend in TCO time series is therefore probably due also to an increase in and long-range transport of pollutants (mainly from industrial and biomass burning activity) in the free troposphere of the southern tropics and subtropics, as suggested by previous studies (Diab et al., 2004; Thompson et al., 2014; Clain et al., 2009). Thompson et al. (2014) highlighted significant increases throughout the free troposphere, 20–30 % decade-1 over Irene and up to 50 % decade-1 over Reunion in the southern subtropical region.

Furthermore, Heue et al. (2016) showed that the tropical tropospheric ozone trend is 0.7±0.12 DU (approximately 3.5 %) decade-1 between 1995 and 2015. There results show that there is an improvement in ozone from the ground to the middle atmosphere in the studied regions. Tropospheric ozone affects total column ozone trends; however, the contribution is not sufficient to modify the TCO trend compared to stratospheric ozone, which represents 90 % of the total ozone present.

In this investigation, TCO analysis indicated a positive trend over the study region. However, a trend investigation in terms of vertical ozone distribution is necessary in order to understand trend variation through different altitude levels. Equatorial, tropical and subtropical ozone trend profiles estimated by the Trend-Run model are presented in Fig. 15, panels (a), (b) and (c) respectively. Considering the temporal coverage of Irene ozonesonde data (1998–2008 and some profiles recorded in 2012), the Irene and Reunion ozone profiles were separated for the sake of trend analysis. Figure 15c1 shows the ozone trend profile for Reunion, while that for Irene (January 1998 to December 2008) is shown in panel c2. Overall, the ozone trends obtained are negative in the upper troposphere and positive for altitudes higher than 22 km. This indicates an improvement in stratospheric ozone. For the equatorial region, the trend profile shows a positive gradient from 15 km to the tropopause (approximately 18 km), followed by a maximum positive ozone trend in the 18–22 km layer (about +2 %). A negative trend is seen between 22 and 25 km and a zero trend for heights above 25 km. The tropical trend evolution below 22 km is similar to the trend observed over Reunion (Fig. 15c1), which is marked by a negative peak around 19 km. Such a negative peak is probably a characteristic of the tropical trend and highlighted at Reunion due to the proximity of this station to the tropical zone.

The results in trend estimates over the equatorial region are similar to previous studies based on a combination of GOMOS and SAGE II data recorded between 1984 and 2011 performed by Kyrölä et al. (2013). This work showed an apparent improvement in ozone trends onwards from 1997. The recorded trend (1997–2011) in the 20–30 km altitude range is consistent with results obtained in the present study and can be summarized as follows: a positive trend of approximately 2 % in the altitude range 19–21 km and approximately 0 % between altitudes 21 and 30 km. A trend varying between 0 and 1 % was also observed by Randel and Thompson (2011) in an altitude range of 23 to 35 km by averaging SHADOZ and SAGE II ozone profiles recorded between 1984 and 2009 for latitudes 20∘ N–20∘ S. Furthermore, Randel and Thompson (2011) observed a negative peak at 19 km, similar to what was observed in this study over Reunion and the tropical region. This negative peak is explained as resulting from a systematic increased tropical lower stratospheric upwelling. As reported by the WMO (2014), the increase in tropical upwelling is associated with a strengthening of the Brewer–Dobson circulation caused by GHG-induced climate change. Furthermore, Lamarque and Solomon (2010) associated the declining trend in ozone levels in the lower tropical stratosphere with an increase in carbon dioxide and sea surface temperature. They also demonstrated that a decrease in ozone in the upper stratosphere could be strongly linked to an increase in halogen compounds at high altitudes. The results obtained in this work indicate that the positive trends observed over Reunion and Irene from TCO analyses are probably due to an increase in ozone levels in the stratosphere, which is associated with a decrease in halogen compounds and EESC at high altitudes. Declining EESC levels in the subtropical region constitutes the main reason for ozone increase as a result of the Montreal Protocol and its amendments (Yang et al., 2006; Anderson et al., 2000; Reinsel, 2002; WMO, 2010, 2014).

The positive trend of 2 % in ozone levels recorded over the equatorial region in the 18–22 km altitude range cannot be explained by the decline of EESC compounds in the stratosphere. Instead, this trend could rather be linked to an increase in NOx (NO+NO2) in this altitude range (Gebhardt et al., 2014; Nevison et al., 1999). Furthermore, Gebhardt et al. (2014) obtained a positive ozone trend at altitudes higher than 18 km and the variation pattern observed is similar to that obtained in our study up to an altitude of approximately 23 km. The investigation of Gebhardt et al. (2014) was based on SCIAMACHY satellite observations (2002–2012). They associated the observed positive trend at 18–22 km with an increase in NOx concentration in the lower stratosphere. The idea that NOx levels may have a direct effect on stratospheric ozone variability is not new and was first suggested by Nevison et al. (1999). In this study they used a model in order to identify the NOx contribution to ozone variability at different altitudes in the stratosphere. Through these calculations it was found that the increase in NOx compounds in the lower stratosphere leads to an increase in ozone levels at this altitude. It is important to note that NOx can have a buffering effect on the Clx and HOx ozone-destruction cycles in the tropical lower stratosphere as explained by Nevison et al. (1999).