The stratospheric ozone layer protects the biosphere from harmful UV radiation. One of the important measures that regulate the amount of UV radiation reaching the surface is the total column amount of ozone or, in short, total ozone, which is defined by the vertical integration of the ozone number density profile. As the ozone profile peaks in the lower stratosphere, total ozone is also representative of lower stratospheric ozone (from tropopause to about 27 km). The strong decline in global total ozone observed throughout the 1980s and the discovery of the Antarctic ozone hole raised the awareness for the need to protect the ozone layer that culminated in the 1985 Vienna Convention to take action. The main cause for the severe ozone depletion was identified as halogen-containing substances also called ozone-depleting substances (ODSs), that are sufficiently long-lived to reach the stratosphere, releasing halogens that destroy ozone e.g.,. The Montreal Protocol and its amendments which were initiated in 1986 became a binding agreement on phasing out ODSs, that ultimately initiated a decline in stratospheric halogens about 10 years later e.g.,.

Satellite and ground-based data revealed a dramatic total ozone column decline of about -3 to -6 %decade-1 (dependent on latitude) throughout the 1980s until the mid-1990s that was linked to observed ODS increases and references therein. In the Northern Hemisphere (NH), the lowest annual mean total column ozone levels occurred in 1993, resulting from enhanced stratospheric aerosol-related ozone loss after the major volcanic eruption of Mt. Pinatubo in 1991 a few years before the peak in stratospheric ODSs was reached e.g.,. In the late 1990s, annual mean total ozone increased rapidly in the NH, faster than expected from the slow decrease in ODSs as a result of measures taken in response to the Montreal Protocol and its amendments. This rapid increase in the NH revealed the important role of atmospheric dynamics, notably ozone transport via the Brewer–Dobson circulation that causes large variability on interannual and intra-annual timescales e.g.,.

Apart from the interannual variability, total ozone levels have remained globally stable since about the year 2000. The success of the Montreal Protocol agreement is thus undisputed as the earlier decline in total ozone was successfully stopped . Since ODS levels (outside of the polar regions) are expected to decrease slowly at about one-third of the absolute rate of the earlier ODS increase see Fig. 2 in, it is expected that the onset of ozone recovery should be evident. There are two possible explanations as to why this has not been observed globally yet. Positive ozone trends are too small to be detected relative to the observed large variability and, secondly, ODS-related ozone trends are in competition with trends due to climate feedbacks. The latter means total ozone trends are not necessarily congruent with stratospheric halogen trends; e.g., they have the same ratio of trends before and after the ODS peak as ODSs themselves. For instance, the observed increase of upper stratospheric ozone (∼ 2 hPa) of about 2–4 % per decade since 2000 had about equal contributions from climate change and ODS changes as deduced from chemistry–climate models see Figs. 2–20 and related references in.

Regular stratospheric ozone observations started with ground-based Dobson spectrophotometers in the mid-1920s . The number of stations with regular Dobson spectrophotometer observations strongly increased after the International Geophysical Year (IPY) 1957/1958 . First measurements of ozone from space occurred in 1970 with the launch of the BUV (Backscatter UV) spectrometer. Continuous measurements from space started at the end of 1978 with the Solar Backscatter UltraViolet (SBUV) and Total Ozone Mapping Spectrometer (TOMS) instruments . Starting in 1995, the SBUV-2 and TOMS observations were complemented by the European Global Ozone Monitoring Experiment (GOME)-type instruments that in addition to ozone measure other important species (NO2 and OClO) relevant for stratospheric ozone chemistry e.g.,.

Global and continuous ozone observations from space now span a time period of nearly 40 years. These observations now extend to about 20 years after the global stratospheric ODS peak occurring in approximately 1996 (or 16 years after the later ODS peak in polar regions). This is near the minimum number of years of observations required to obtain statistically significant ozone trends in the absence of other competing processes contributing to long-term ozone changes .

This paper reports on updated total ozone trends by adding four more years of data (2013–2016) compared to results presented in the last World Meteorological Organization (WMO) ozone assessment . As most satellite instruments have a limited lifetime of generally less than 10 years, long-term trends can only be investigated by using merged datasets. Currently, there are four different satellite datasets available; two of them rely on the series of SBUV instruments covering the period since 1979 and two datasets combine the European UV nadir sounders (GOME, GOME-2, OMI, SCIAMACHY) starting from 1995 . These satellite datasets are complemented by a fifth dataset that is based on monthly mean zonal mean total ozone data derived from ground-based UV spectrometer data, mainly Dobsons and Brewers, which are collected at the WOUDC (World Ozone and UV Database Center) at Environment and Climate Change Canada . The regression analysis applied to these data is similar to that described in and focuses on annual mean zonal mean data. The main difference to the earlier study is that we use in this paper five merged datasets, while in only the GSG and SBUV MOD datasets were used. All datasets used here were updated up to and including 2016 (four more years added). In , the piecewise linear trends (PLTs) and equivalent effective stratospheric chlorine (EESC) term were fitted, while here only the independent linear trends (ILTs) before and after the turnaround in ODSs are considered for the reasons discussed in Sect. .

In Sect. , the five merged datasets are briefly described and followed in Sect.  by a description of the multiple linear regression (MLR) used in the trend analysis. Section  shows the results of total ozone trends in rather broad zonal bands (Southern Hemisphere and Northern Hemisphere extratropics and tropics) that are commonly used for ozone profile trends . This will allow us to look at the consistency between lower stratospheric ozone (derived from profile observations) and total ozone trends. In Sect. , latitude-dependent annual mean trends are presented and discussed. Results will be also shown in Sect.  for selected months during polar spring as recovery of Antarctic ozone levels in September has been recently reported by . A summary and final remarks are given in Sect. .

A total of five merged and homogenized datasets are used in this study. There are two different versions of merged datasets from the series of SBUV and SBUV-2 satellite instruments (NASA SBUV MOD v8.6 and National Oceanic and Atmospheric Administration (NOAA) SBUV Merge v8.6) that have been operated continuously since the late 1970s. Two merged datasets are mainly based on the series of European satellite spectrometers GOME, Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY), and GOME-2A which use different retrieval algorithms and slightly different merging approaches (University of Bremen GSG and ESA/DLR GTO datasets). Both datasets cover the period from 1995 to today. The fifth dataset is the monthly mean zonal mean data from the network of ground-based Brewers, Dobsons, SAOZ (Système d'Analyse par Observations Zénithales), and filter instruments collected at the WOUDC . The data sources are summarized in Table  and the various datasets are briefly described in the following subsections.

In this section, the MLR equation and the various explanatory variables used are briefly summarized (Sect. ), followed by a discussion on the various choices of trend terms, e.g., independent linear trends before and after the turnaround of the stratospheric halogen (preferred choice in this study), hockey stick, or EESC curve (Sect. ).

In Figs.  and , the five bias-corrected merged time series are shown for the extratropical 35–60∘ zonal bands in the Northern Hemisphere and Southern Hemisphere, respectively. In the NH, the results from the MLR are only shown for the NOAA dataset and are indicated by the orange line. In the SH, the MLR results from the WOUDC data are indicated. In general, the agreement between the datasets are better than those with the MLR results, but also the MLR works reasonably well, explaining about 85 % of the variance in the time series. There is overall a high consistency between all datasets in the extratropics. The standard MLR plus AO and NH BDC terms were used in the NH, while in the SH only the SH BDC term was added.

Before 1997, total ozone trends in the extratropical belts between 35 and 60∘ in each hemisphere were about -3±1.5(2σ) %decade-1. The trends changed to about zero to +0.5 %decade-1 after the ODS peak in the extratropics. The recent trends are mostly statistically not significantly different from a zero trend, meaning total ozone levels remained stable in the extratropics over the last 20 years (1996–2016). Nevertheless, the trend change is significant and it confirms the conclusions from the last WMO ozone assessment that the ODS-related decline was successfully stopped .

Table summarizes the post-ODS peak trends for the five datasets considered here. In the NH extratropics, most data show a near-zero trend. In the SH extratropics, trends are positive and slightly larger than in the NH. The GSG, GTO, and WOUDC datasets indicate a positive trend of 0.7 %decade-1 here, barely reaching the 2σ uncertainty level. Except for the NASA dataset, all datasets show a positive trend of +0.5 %decade-1 or more in the SH.

Figures  and (NH and SH, respectively) show how the post-ODS peak trend changed during the last decade by adding more years of observations since 2006. Up until 2010, the linear trends in the NH were at about +1 %decade-1 with an uncertainty just less than 2 % (2σ). With additional years after 2010, trends lowered to about +0.5 %decade-1. The uncertainty is now reduced to slightly below 1 %decade-1. This means that a trend of 1 %decade-1 could be observed after 20 years of observations following the ODS peak. The below-average annual mean NH total ozone in 2016 is linked to the severe Arctic ozone depletion in the same year and related to the anomalous QBO-induced meridional circulation changes . This resulted in a drop of the 1997–2016 NH ozone trend down to +0.4 %decade-1 (compared to +0.6 %decade-1 ending in 2015). The trend estimates are somewhat dependent on the end value in the time series. In 2010, NH extratropical ozone levels were unusually high (see Fig.  and ). Despite the reasonable fitting, this high anomaly increased the trend through 2010 to +1.8 %decade-1 which was statistically significant at that time (Fig. ).

The trend results do not vary much with additional terms used in the MLR. The standard MLR and the extended MLR (adding BDC-N and AO in the NH and BDC-S in the SH) yield about the same trend results, but the latter provides smaller uncertainties because the explained variance increases significantly with the added terms (∼ 10 % in the NH). In the SH extratropics (Fig. ), the trends did not vary much during the last few years, but uncertainties have been reduced to slightly below 1 %decade-1.

In the tropics, both GSG and WOUDC show significant trends of +0.8±0.4 and +0.5±0.5 %decade-1 after 1996, respectively, while all other datasets (NASA, NOAA, GTO) show smaller and insignificant trends (Table  and Fig. ). It appears that for the former datasets, in particular the GSG dataset, some decadal drifts are evident. The difference between the maximum and lowest trends is less than 1 %decade-1 which is within the 1–3 %decade-1 stability requirement for long-term satellite datasets .

One should keep in mind that significance of trends in some zonal bands and for some datasets that are barely significant at 2σ can easily vanish depending on the choice of proxies or set of fitting parameters . Given the fact that additional uncertainties from the merging of the datasets as well as in the calculation of zonal mean data from sparse ground-based data are not accounted for here, all observed trends are likely not significant yet.

In the last ozone assessment , a near-global average (60∘ S–60∘ N) increase of about +1±1.7 % from ground and space measurements from 2000 to 2013 (corresponding roughly to a 0.8 %decade-1 increase) was reported. For the extended period considered here (1997–2016), the trends appear much smaller (near-zero trends in the tropics and NH, except for two datasets in the tropics). Only in the SH the trends are about 0.6±0.6 %decade-1 for most datasets (see Table ). In the extratropics, trends (Figs.  and ) were reduced by about half by extending the time series from 2013 to 2016, although this difference is within the trend uncertainties. It is evident from the time series (Figs.  and ) that most of the added years since 2013 show below-average ozone compared to the decade before.

The pre-ODS peak trends derived here are in good agreement with the integrated profile trends reported in Tables 2–4 of . The trends after 1997 reported here are about half of the trends reported by , as explained above. Nevertheless, within the combined uncertainties, trends agree. Some of the differences may also be due to the different time periods considered (e.g., starting in 2000 versus 1997).

Our results are also largely consistent with more recent profile trend studies that basically show mostly insignificant trends at lower stratosphere altitudes.

In Fig. , zonal mean total ozone trends before and after the ODS peak in 1996 are shown for all five datasets as a function of latitude from 60∘ S to 60∘ N in steps of 5∘. In order to better compare the results from one dataset to the others, all remaining datasets are overplotted without their uncertainties. For all datasets, the trends since 1996 are mostly below 1 %decade-1 similar to the results obtained in our previous study and what was derived from the broader zonal bands (previous section). For some latitudes, trends are barely statistically significant at 2σ. Before discussing the trends in more detail, the way the MLR was applied to obtain the trends as well as some other diagnostics will be presented and discussed.

The trends were calculated using the full MLR. The regression at each latitude band was repeated by removing those terms in the extended regression (Eq. ) for which the corresponding fit coefficient was smaller than its 2σ uncertainty. Figure  shows the square correlation between the regression model and observation and χ values as a function of latitude for the NASA and NOAA regressions. The square correlation varies between 0.7 and 0.9 for the full regression. The results for the NASA fit using the standard regression are also shown, demonstrating that adding the BDC-S term improves the fits at the SH middle latitudes and NH tropics (higher r2 and lower χ), while BDC-N and AO improve at NH middle latitudes. In the SH low latitudes, the standard model was sufficient (no additional terms needed). The importance of the BDC-S term in the NH tropics is for the first time reported and will be discussed later.

An important question arises as to how sensitive the trends are, in particular the ones after 1996, to additional terms from Eq. () in the regression. As an example, the trend results for the NOAA data using the standard model and the full MLR are displayed in Fig. . The post-ODS peak trends are nearly unchanged, indicating that the recent trends are not sensitive to the additional terms used, which is the case for all datasets; however, the full MLR reduces the trend uncertainty. Within the uncertainties, the pre-1996 trends are also identical in the standard and full MLRs. At the NH middle latitudes, the addition of the BDC-N and AO terms reduces the downward trend until 1996 by about 1 %decade-1. As all proxies were not detrended, the AO and BDC-N terms also contribute to the long-term trends (thus reducing the remaining linear trends). Apart from the year-to-year variability, the AO index increased throughout the 1980s along with the EESC (ODSs) as shown in Fig. 1 in (see also ). The very high total ozone observed at NH middle latitudes in 2010 (Fig. ) was linked to extreme negative AO as well as a very strong NH BDC during Arctic winter in the same year.

The contribution of the various factors (solar cycle, QBO, ENSO, aerosol, and so on) to ozone variability as a function of latitude is shown in Fig.  for two of the datasets (NASA, WOUDC). Plotted are the signed maximum responses in Dobson units (DU), which are the differences between the maximum and minimum values of the regression term time series. A negative sign means that the ozone response is anticorrelated with the proxy change. The ozone response to the factors are in very good agreement with our previous results from based on data up to 2012. The maximum solar response of about 4–6 DU in the tropics is in agreement with the ∼2 % change from solar minimum to maximum in the lower stratosphere reported by . Solar ozone responses are significant at all latitudes and are the result of the solar impact on atmospheric dynamics .

In the inner tropics, the ozone response to the QBO terms changes sign poleward of 10–15∘ latitudes in each hemisphere, which means positive ozone changes in the inner tropics are observed in years dominated by the QBO west phase. A new result is that the BDC-S has a significant contribution at low NH latitudes. At middle latitudes above about 40∘, ozone increases are associated with high absolute eddy heat fluxes (BDC proxy) as expected from the enhanced downwelling related to a stronger residual circulation. The opposite effect is seen at low latitudes (ascending branch of the BDC) with lower ozone due to enhanced upwelling and horizontal divergence . Indeed, the BDC-S ozone response has opposite signs between the low and high latitudes. The extension of the BDC-S response into NH low latitudes may be a result of the upper branch of the SH meridional circulation extending into the NH . It is somewhat surprising that a similar tropical response is not evident in the NH. However, the QBO indices have a significant correlation with the BDC-N proxy (r∼-0.7). The lower stratospheric QBO in the west phase (positive QBO index) allows planetary waves to be more strongly deflected towards the Equator, thus reducing the perturbation of the westerly flow in the extratropical stratosphere , resulting in a weakening of the meridional winter BDC, lower middle latitude eddy heat flux, and reduced high latitude ozone due to reduced downwelling and higher ozone losses due to lower polar stratospheric temperatures e.g.,.

The aerosol effect due to the Mt. Pinatubo eruption in 1991 has the largest effect on ozone at high northern latitudes with a reduction of up to 20 DU (NASA) to 25 DU (WOUDC) in 1993. Significant ozone depletion was also observed in the NH following the El Chichón major volcanic eruption in 1982 e.g.,. A positive ozone response to the El Chichón is evident in the SH middle latitudes, most likely due to the specific circulation changes induced by this volcanic event . This is also believed to have caused an initial extratropical increase in SH extratropical total ozone during the first 6 months following the Pinatubo eruption.

Similar to the results from the broad zonal band trends, Fig.  shows that the latitude-dependent post-ODS peak trends (Fig. ) are generally smaller than the trends reported in the last WMO/UNEP ozone assessment which varied between +1 and +2 %decade-1. The NH extratropical trends are below +0.5 %decade-1 and statistically insignificant. In the SH, trends can reach up to +0.7 %decade-1 and at some latitudes barely reach the 2σ uncertainty level, except for the NASA dataset.

Largest variations in trends between the datasets are seen in the tropics. Here, both SBUV datasets show basically zero trends, the WOUDC and GTO negative trends in the inner tropics, and GSG statistically significant positive trends that are near 10∘ latitudes reaching about +0.8 %decade-1. Near the same latitudes, WOUDC trends are also positive and statistically significant. One large issue is that the ground-based data are quite sparse in the tropics, particularly at SH latitudes, and generally towards the end of the data record as many stations have not yet submitted updates to the database.

An interesting result is that NH subtropical trends (20–30∘ N) peak at about +1 %decade-1 and are significant, with the exception of those in the GTO dataset, which are at the lower end of the range observed. The subtropics are regions where total ozone shows quite large gradients in the transition from the tropics (lower ozone) to the extratropics (higher ozone). A shift of the subtropical transport barrier into the tropical region could increase ozone at subtropical latitudes. Indeed, a southward shift of about 5∘ of the tropical belt below 30 km altitude has been inferred from lower stratospheric ozone trends . A recent study by indicates that lower stratospheric age of air in the NH subtropics and extratropics has been increasing in recent years (subtropical air becoming more extratropical and reduced BDC in NH), while in the SH subtropics age of air has variable trends in the lower stratosphere that can be negative and positive depending on altitude and is largely negative in the SH extratropics. The latter would mean that the BDC is getting stronger in the SH, which would result in larger SH extratropical lower stratospheric ozone trends as compared to the NH. However, the recent stratospheric ozone profile trend studies do not indicate such a hemispheric trend asymmetry in the lower stratosphere .

In a recent study by , evidence for a significant positive trend in the SH polar region in September was reported. Other studies also indicated some early signs of ozone recovery in Antarctic spring and summer . September and October are months when the ozone hole area reaches its maximum and total ozone above Antarctica exhibits minimum values (see https://ozonewatch.gsfc.nasa.gov/meteorology/SH.html). A MLR has been applied to monthly mean polar total ozone for September and October in the SH as well as March in the NH. In the Arctic, substantial polar ozone depletion is sporadically observed when stratospheric winter and spring are sufficiently cold e.g.,. For these 3 months, the monthly mean proxies for the respective months were used in the MLR, except for the BDC proxies which were taken as an average from March to September or October in the SH, respectively, and from September to March in the NH. We use the year 2000 as a start for the post-ODS peak trends . The regression results are summarized in Table  and MLR time series are shown for each of the months for one of the total ozone datasets in Fig. .

In SH September, the post-ODS peak trends of the various datasets vary between +8 and +10 %decade-1 with a 2σ uncertainty of about 7 %decade-1. The Antarctic September trend is barely significant at the 2σ level and confirms the findings of . Changes in the regression model, use of different proxies, and considerations of inherent drift uncertainties can easily remove the significance . In contrast, the October trends are much smaller (about 3 %decade-1) and statistically insignificant, which is also in agreement with .

and showed from chemistry–climate model simulations that the Calbuco volcanic event substantially contributed to the observed polar ozone loss in 2015. Even though we used the aerosol data from , as used by and as input to their climate model, as a proxy in our regression, the impact of the aerosol term was found to be negligible in 2015. The apparent contradiction of the aerosol impact on Antarctic ozone between Solomon and Ivy et al. and our study should not be overstated. The fitting of the aerosol proxy data based on Mills et al. is dominated by the Pinatubo event and may therefore not be properly scaled during the Calbuco volcanic event. It is difficult to isolate minor volcanic events with stratospheric impact in the MLR using separate aerosol proxy terms as is done for the larger El Chichón and Pinatubo events. This is clearly a limitation of the MLR approach.

In the Arctic, March total ozone trends are quite small (below 1 %decade-1) and insignificant, similar to the trends observed in NH middle latitude annual means, albeit with much larger uncertainties (on the order of 4 %decade-1 at 2σ). Also, the pre-ODS peak trends in the Arctic (about -3 %decade-1) are similar to the annual mean trends observed in the extratropics (30–60∘ N). At first sight, it seems surprising, as in the 1990s and selected years after 2000 there was substantial polar ozone depletion. As the polar ozone losses occur mostly in cold Arctic winters that are usually associated with years of very low BDC driving, it seems that the BDC term in the MLR accounted for the polar chemical losses. The remaining trends are in excellent agreement with the “gas-phase” chemistry trends at middle latitudes (before and after the ODS peak). In the SH, the polar ozone losses are much larger and the “linear” scaling of polar losses with the BDC proxy is not fully given so that the Antarctic trends are larger, or, in other words, the linear trends may have non-negligible contributions from polar ozone losses.

Updated trends were derived from five different merged total ozone datasets that have been extended up to the year 2016. A MLR with ILTs before and after the maximum stratospheric halogen content (∼ 1996) was applied to annual means in broad zonal bands as well as narrow 5∘ latitude bands up to 60∘ latitudes. In most cases, the results from the last ozone assessment and from other studies and earlier studies confirmed that total ozone has been stable since about 1996, which is a significant change from the earlier decline observed globally outside the tropics. Globally, the post-ODS peak trends vary generally between near-zero trends (NH extratropics) and positive trends of +0.7 %decade-1 (SH extratropics) with a statistical trend uncertainty of about 0.7 %decade-1 (2σ) after 20 years of observations. We may therefore conclude that we are about to emerge into the phase of ozone recovery as is also shown by chemistry–climate and chemistry–transport models e.g.,. Both the regression applied to datasets (e.g., in our study) and models capture the dynamical variability well and their results are consistent.

All post-ODS peak trends are about half of the trends reported in but the changes are still within the trend uncertainties. The main reason is that in most regions total ozone in recent years showed annual means that were lower than the recent decadal mean but were well within the variability that was observed during the last 20 years.

In some regions, some of the datasets show significant positive trends. In the tropical band (<20∘), recent trends are significant for two (GSG, WOUDC) and in the SH (35–60∘ S) for three (GSG, GTO, WOUDC) out of five datasets (Table ). The significance of these trend estimates is close to 2σ. The uncertainties reported here are purely statistical and do not account for uncertainties that may arise from the merging of the individual satellites as well as from sparse sampling of ground-based data affecting the zonal mean estimates. Also, the significance of trends may get altered (or become insignificant) depending on the explicit choice of regression setup (e.g., which terms to add) as well as choice of proxies for a given process. The latitude-dependent trends (Fig. ) after 1996 are largely consistent with the results from the broader zonal bands. A striking feature is that most datasets see larger positive and statistically significant trends at subtropical latitudes between 20 and 30∘ N. A southward shift of the tropical belt e.g., could be a potential explanation; however, a recent study shows that a markedly positive trend is not observed in most ozone profile datasets .

The higher trends at NH subtropics have some impact on the near-global trends (60∘ S–60∘ N) derived from our MLR analyses as summarized in Table  and Fig. . Three out of the five datasets (NOAA, GSG, and WOUDC) show statistically significant trends of about +0.6±0.3 %decade-1 on average. This trend is smaller than the trend derived from profile data for the period 2000 to 2013 (+1.1±1.7 %decade-1) reported in Tables 2–4 of which was derived from the combination of ozone profile data. Figure  shows the MLR results of data having the lowest (GTO) and highest post-ODS peak trends (GSG). One should keep in mind that from MLR analyses alone we can not uniquely attribute the observed trends, as they may have a significant contribution from climate change and possible feedback on atmospheric dynamics and chemistry that are difficult to disentangle without the use of chemistry–climate models.

The observed positive trends above Antarctica in September since 2000 as reported by were confirmed by our MLR analysis; however, the impact from aersols from the recent series of minor volcanic eruptions was found to be minor in contrast to the results from and . In October, the MLR trends above Antarctica were much smaller and statistically not different from zero as were trends from the Arctic in March for all five datasets.

Adding 4 years of data in the various long-term total ozone data records has now further reduced the statistical uncertainties in the zonal mean trends to below 1 %decade-1. We consider the uncertainties cited here as lower limits, as we do not account for added uncertainties from the drifts in and from merging the data, the latter needed to obtain long-term datasets, and the data sampling was low (mainly ground-based data).

Continued ozone observations and monitoring are needed to consolidate the evidence of ozone recovery and also further improve our understanding of the complex ozone–climate feedback (in combination with chemistry–climate modeling) that will have a significant impact on future evolution of ozone .