Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
The zonally averaged total column ozone from ground-based observations for the period 1964–2008 is being examined to detect changes in the trend pattern. The ozone long-term changes for the periods 1980–1995 and 1996–2008 are estimated through the use of a triad of regression models that differ in the description of a trend term. The trend term could be proportional to the amount of ozone-depleting substances in the atmosphere, or has a piecewise linear form with turning points in 1980 and 1996, or described by any smooth curve. The standard indices of the ozone dynamical drivers: 10.7 cm solar flux, zonal component of the stratospheric wind in the tropics, the Arctic, and Antarctic Oscillation index, and the vertical component of the Eliassen-Palm flux are used to parameterise the ozone response to changes in the atmospheric dynamics.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Stratospheric ozone issue has focused on scientific and public interest because of its climate impacts and role in shielding the Earth from UV solar radiation. The total column ozone depletion levelling off began in the mid-1990s over the extratropical regions after a period of two decades with a strong ozone decline. It has been linked to the success of the Montreal Protocol for controlling production of ozone depleting substances (ODSs), WMO (2007).
Many attempts to document the decline and subsequent recovery of the stratospheric ozone pointed to the uncertainty in estimation of the anthropogenic trend component due to the significant year-to-year fluctuations in ozone driven by the interannual variability of the atmosphere dynamics (e.g. Randel et al., 2002; Reinsel et al., 2005; Hood and Soukharev, 2005; Austin and Wilson, 2006; Dameris et al., 2006; Dhomse et al., 2006; Weatherhead and Andersen, 2006; Harris et al., 2008; Peters et al., 2008). Climate changes influencing the atmosphere circulation and chemistry make attribution of the ozone changes to anthropogenic effects even more difficult. Increased greenhouse gases (GHGs) induce cooling of the stratosphere that leads to an increase in the upper stratospheric ozone due to slowing the ozone loss reaction in lower temperature (Barnett et al., 1975; Jonsson et al., 2004). Simulations from coupled chemistry-climate models show that the Brewer-Dobson circulation, which redistributes ozone from the source region in the tropics to mid- and high latitudes, will speed up with the increase of GHGs (e.g. Butchart et al., 2006). An accelerated Brewer-Dobson circulation removes more ozone from the tropics to the mid-and high latitudes inducing a decline of ozone in the tropics and the compensating increase in the extratropics (Austin and Wilson, 2006; Shepherd, 2008; Waugh et al., 2009).
WMO (2007) has defined the following stages of the ozone recovery: (1) a slowing in the downward trend of the ozone; (2) onset of the upward trend in the ozone time series having ‘natural’ variations removed; and (3) returning to pre-1980 ozone levels in the stratosphere. Recent WMO reports on the state of ozone layer (WMO, 2003, WMO, 2007) corroborated that the increase of the atmosphere contamination by ODSs was mostly responsible for the ozone depletion in the extratropics in the 1980s and early 1990s. As the concentration of ODSs started to decline since mid-1990s more efforts have been paid to search for the first signs of the ozone recovery. Ozone variations due to natural processes are superposed on the anthropogenic ones and they are involved in complex relationship, e.g. the ozone destruction is more effective after large volcanic eruptions in period when the stratosphere is polluted by ODSs. Substantial year-to-year ozone variability reflecting the atmospheric dynamics/chemistry changes hampers to identify a weak trend (as it is expected since the mid-1990s) in the data.
Vyushin et al. (2007) examined the merged satellite total ozone data 1979–2005 for the trend variability in the period 1979–1995 and 1996–2005. The trends (in monthly mean data for 5° latitudinal zones) were mostly positive in the period 1996–2005 but statistically significant only in the narrow band in the NH midlatitudes. Yang et al. (2009) reported the statistically significant positive TO3 trend of about 1% per decade in the period 1996–2007 for the averaged 50°S–50°N satellite data. Angell and Free (2009) found the global trend of 0.7% ± 0.5% (2σ) per decade at the end of time series (beginning of the 2000s) and the upward trends were also identified over the north temperate and Tropics zone.
The main objective of this study is to identify the regions of the globe and periods of the year with a statistically significant upward trend in the zonally averaged ground-based total ozone (TO3) data for the period 1996–2008. We analyse the performance of various statistical trend models being used in recent attempts of the trend calculations for the period of strong increase (1980–1995) and slow decrease (1996–2008) in the ODSs concentration in the stratosphere. Effects of various parameterisations of the dynamically driven ozone variability on the trend values are also examined. If all examined trend models provide statistically significant positive trends since the mid-1990s for the specific zone and the season of the year it will strongly support the case for the second stage of ozone layer recovery.
2. Total ozone data
The individual station TO3 monthly means derived from ground-based measurements by the Dobson and Brewer spectrophotometers, and Russian filter ozonometers, which have been deposited at the World Ozone and Ultraviolet Data Centre (WOUDC), are averaged over zonal bands. The monthly mean data for 5° zones for the period January 1964 to December 2008 are obtained from WOUDC (available on web page http://www.woudc.org/data_e.html in section “Zonal Mean Ozone from Ground-Based Instruments”) and these are used to form time series for the following broad zones: SH polar (90°S–65°S), SH midlatitudes (60°S–30°S), tropics (25°S–25°N), NH midlatitudes (30°N–60°N), NH polar (65°N–90°N), 50°S–50°N band, and the whole globe (90°S–90°N). The area-weighted zonal monthly means for 5° zones are used to calculate the aggregated zonal means.
WOUDC archives the data submitted by different national agencies using their own measurement practices, data processing, calibration procedures, etc. Thus, the estimation of the precision of ozone ground-based measurements and instrumental trends in the data are not straightforward. Since instrumental errors are typically independent for individual ground-based instruments, ozone values averaged over several stations should have smaller uncertainties. A comparison with the satellite data has shown that the majority of the Dobson and Brewer stations provided reliable results, especially those within the 60°S–60°N band (Fioletov et al., 2008).
3. General concept of trend model
The interannual fluctuations in the atmospheric ozone are a superposition of components due to changes in the atmospheric chemistry (mostly related to changes in ODSs) and those induced by the dynamical processes affecting the ozone layer. In order to study sources of long-term variability in total ozone, it is necessary to separate dynamically driven variations from chemical ones. The regression models used have the general form:
- ΔO3(t) is the ozone fractional deviation in running month t,
t = (year–1964)°12 + M, t = 1 in January 1964, O3(t) is TO3 monthly mean in month t, O3*(M) is the long-term (1964–1979) TO3 monthly mean for calendar month M corresponding to t,
- Trend (t) represents a slowly varying component (trend) of the ozone time series that is assumed to be related to the long-term changes in the anthropogenic forcing on the ozone layer,
- Atmosphere_Dynamics (t) describes part of the ozone variations due to the atmospheric dynamics changes. We consider standard proxies to parameterise dynamically driven signal in ozone,
- Noise (t) represents the noise term that can be partially linked to the presently unknown short-term forcing yielding some autocorrelations in the noise term (usually with a one-month lag).
Model (1) is run separately using all available monthly means (the entire year model) and seasonal subsets of the data (the seasonal model). The winter subset comprises the DJM monthly means (i.e. three monthly values per each year) and, correspondingly, we build the spring (MAM), summer (JJA), and autumn (SON) subset of the data. In our search for the stage of ozone layer recovery we compare the ozone trends in the period 1979–1995 and 1996–2008 using various trend terms of (1) and sets of the proxies.
4. Dynamical proxies
The 10.7 cm solar flux reaching the Earth's surface and zonal wind in the equatorial stratosphere are the ozone explanatory variables frequently used in the previous trend models. These are the best suited ozone drivers because of their narrow spectrum and true ‘natural’ origin of variability being independent of ozone changes (WMO, 2007; Vyushin et al., 2007). Other proxies considered here have been examined in the previous trend models, e.g. the Eliassen-Palm flux by Harris et al. (2008), the Artic Oscillation and the Antarctic Oscillation index by Yang et al. (2009). The following proxies for the dynamical ozone drivers are taken into account:
a)the zonal component of wind at 30 and 50 hPa over the tropics to describe the Quasi-biennial oscillations (QBO). Two wind components are chosen to account for time lag between zonal equatorial wind and its response out of the tropics. Freie Universitat Berlin provides QBO index based on the averaged wind velocity measurements from radiosondings over three equatorial weather stations. Data available on: http://www.geo.fu-berlin.de/met/ag/strat/produkte/qbo/qbo.dat
There is a serious problem with an application of the trend model containing the proxies other than the QBO wind and 10.7 cm solar flux. Such proxies could have their own trends (or wide spectrum of the long-term variability) that may induce erroneous estimation of the anthropogenic trend component. In longer time scales the ozone changes could induce disturbances in the atmospheric dynamics possibly affecting its explanatory variables. The relationship between ozone and its regressor established for shorter time scale variations, when the correlated changes in ozone and the proxies reflect common dynamical process affecting both variables, cannot be automatically extrapolated to longer time scales. It cannot be excluded that the long-term variability of the AO (or AAO) index, and the EP flux proxy are somewhat affected by the climate changes partially induced by the ozone changes. Evidently these proxies are not perfect regressors, but it will be interesting to see how ozone layer recovery will look if more ozone variability is potentially attributed to the dynamical forcing.
5. Trend term
We examine various versions of model (1) with a different description of the trend term and number of the dynamical proxies. The following options of the trend component are considered:
- EESC model
where EESC(t) is time series of the ODSs concentration in the stratosphere parameterised by amount of Equivalent Effective Stratospheric Chlorine (EESC). This quantity combines the destructive power of all the chlorine and bromine containing species with weighting corresponding to their individual ozone-depleting potentials. t1 is selected as the beginning of the ozone series, January 1964. The regression constant, αEESC, could depend on season. The EESC time series is taken from the European Environment Agency web site http://dataservice.eea.europa.eu/dataservice/metadetails.asp?id = 999
- PWL model
where the trend component is described as a piecewise linear (PWL) function with 3 joint straight lines with turning points: January 1980 (the onset of ozone decline) and January 1996 (the onset of the ozone recovery), εt, 1980 = 1 if t ≥ 1980 and 0 otherwise, εt, 1996 = 1 if t ≥ 1996 and 0 otherwise. All regression coefficients are allowed to be seasonally dependent, βPWL and βPWL + γPWL represent the rate of ozone change (%/per decade) in the period 1980–1995 and 1996–2008, respectively. Selection of the turning points is similar to that used in recent trend models (e.g. Vyushin et al., 2007; Yang et al., 2009).
- FLEX model
were Smooth (t) is a smooth function representing the long-term component of the TO3 time series that is filtered out from the residual time series that remains after subtracting dynamically driven variations from the original series. This model is called flexible (FLEX) trend model as a shape of the trend curve is not a priori defined but it is extracted from the time series of residuals.
The trend models examined here have been extensively used in previous estimates of the long-term ozone changes: EESC model (WMO, 2007; Harriset al., 2008), PWL (Reinsel et al., 2005; Miller et al., 2006; Vyushin et al., 2007; Yang et al., 2009), and FLEX (Harris et al., 2001; Krzyścin, 2006; Krzyścin and Rajewska-Wiȩch, 2009).
Figure 1 illustrates the trend patterns retrieved from all monthly means for the 50°S–50°N zone. The trend curves are vertically shifted to have zero value at the beginning of the time series. The trend patterns by the different models coincide, but that is not always the case for the other zonal data.
None of the models prevailed over the others. The limitation of PWL is a prescribed time of the turning point and assumption of a sharp turnaround from decreasing to increasing phase of the trend. FLEX allows more freedom for selecting the trend pattern. However, the type of data smoother and the smoothness level are arbitrarily chosen and need additional model runs to select the best option (Krzyścin and Borkowski, 2008). We apply the Lowess smoother to disclose a smooth curve hidden in the models' residuals, i.e. in original data minus the parameterised ‘natural’ variations. The smooth curve is therefore interpreted as the anthropogenic trend component.
The assumption that the trend pattern follows changes in ODSs concentration in the atmosphere (EESC model) allows direct calculation of the anthropogenic part of the ozone long-term change. PWL and FLEX models provide the trend component comprising, besides the anthropogenic signal due to the ODSs variability, also other long-term fluctuations forced by unknown yet phenomena and processes not parameterised by the trend model. For example, the atmospheric circulation changes were induced by the lowering of the stratospheric temperature due to the GHGs increase in the troposphere, and such changes could affect the ozone global distribution.
The pattern of the ozone trend, forced by the atmospheric chemistry changes, could not exactly follow that of the EESC concentration. Other processes, for example, heterogeneous chemistry on sulfate aerosols and/or on particles of Polar Stratospheric Clouds (PSC) are also involved in the ozone chemical destruction but such effects are not parameterised by EESC model. Moreover, there is the possibility of defining various EESC series depending on the prescribed mean age-of-air (Newman et al., 2007).
To discuss ozone changes in recent years we calculate the linear trends between 1980 and 1995 (the increasing phase of EESC), and the linear trends between 1996 and 2008 for the period of an expected slowdown of TO3 decline or even the recovery anticipated from the decreasing phase of EESC. The linear trend values are calculated directly by PWL model. These are the regression constants of the model, i.e. slopes of the straight lines. For the other models we calculate the linear equivalent of the calculated ozone change, i.e. the trends are expressed in % per decade after the division of the trend component change over the selected time interval by the duration of the interval.
For all the models examined here, the regression constants are calculated by the penalised least-squares method using a multivariate adaptive regression splines (MARS) technique that allows a dependence of the regression coefficient on value of the pertaining regressor (Krzyścin et al., 2005; Krzyścin, 2006).
The 95% confidence ranges for the trend curve are derived independently for each season and the entire year data by the block bootstrapping. Three-month blocks of Noise(t) time series (or 12-month in case of the entire year's data) chosen for a random year are added to the trend term and ‘natural’ variations (EESC and PWL model), or to the trend term only (FLEX model). In this way, a new hypothetical representative of the time series is built. The bootstrap trend component is calculated by the MARS regression (EESC and PWL model) or LOWES smoother (FLEX) applied to the hypothetical time series. The mean trend curve and its confidence intervals are estimated from a sample comprising 1000 hypothetical trend curves. The methodology of calculating bootstrap samples and confidence limits has been described in our previous paper (Krzyścin and Borkowski, 2008).
The trend values and 95% confidence intervals are calculated from the bootstrap sample containing 1000 trend curves. The calculations are done independently for all seasons and for an entire year's data (all the monthly means). All versions of the trend model are run with the QBO and solar proxies using the zonal TO3 monthly means for the period 1964–2008. The linear trend values are compared with those obtained by FLEX model using all possible proxies described in Section 4. Thus, FLEX model is run twice with limited and full number of proxies. FLEX model with all proxies is run for the shorter period, 1979–2008, because the AO (or AAO) and EP proxy are more reliable in this period as being derived from the satellite observations of meteorological parameters. Figure 2 shows the results for boreal winter, spring, summer, and autumn, respectively. Figure 3 presents the results for the entire year. Figures 2 and 3 illustrate the linear trend values in the period 1980–1995 and 1996–2008 by EESC, PWL, and FLEX model containing the QBO and solar proxies, and by FLEX model using all examined proxies.
All models provide almost the same trend values in the period 1980–1995. Thus, it appears that the pattern of the long-term variability of ozone can well be approximated by a straight line. Concentration of ODSs in the atmosphere increases monotonically during that period and the pattern of EESC time series is almost linear. It is not surprising that the largest negative trends are found in the SH polar zone about—20% per decade in austral spring (the Antarctic ozone hole). The smallest negative trends appear in the Tropics at less than 1% per decade.
A much higher diversity of the ozone trend values is revealed in recent years (1996–2008). The EESC increase is rather small in that period inducing a slight upward TO3 trend ∼1% per decade. EESC model yields statistically significant increases in all zones and seasons. An upward TO3 tendency is forced by the EESC's slight decrease. Other models provide mostly statistically insignificant trends in the period 1996–2008. It means that the ozone decline seen in the former period has not being continuing in the latter period, thus, the first sign of ozone layer recovery could be announced.
We would like to find cases when all examined options of model (1) provide statistically significant upward trend in the period 1996–2008. This would provide a strong support for the ozone layer recovery. All models give statistically significant positive trends in boreal spring (globe, Figure 2(b)), in boreal summer (50°S–50°N zone, (Figure 2(c)), and in the entire year data (50°S–50°N zone, Figure 3).
PWL model reveals additional statistically significant positive trends in the period 1996–2008 in boreal winter (globe and 50°S–50°N zone), spring (NH midlatitudes and NH polar zone, and 50°S–50°N zone), summer (NH midlatitudes, and globe), autumn (globe and SH polar zone), and the whole year (the tropics, NH midlatitudes and polar, globe). A statistically significant continuation of the ozone decline is disclosed by PWL model in boreal spring (SH midlatitudes) and summer (SH polar zone). FLEX model with all available proxies supports also the upward trend in boreal winter (the tropics), spring (the tropics, NH midlatitudes, and 50°S–50°N zone), autumn (SH polar zone), and in the whole year data (the tropics, SH and NH midlatitudes). The continuation of the ozone decline is found only in boreal winter over SH midlatitudes. However, the version of FLEX model based only on the QBO and solar proxies does not support such trends.
7. Discussion and conclusions
We present results of the numerical exercise to delineate trends in the zonally averaged total ozone data for the period 1980–1995 and 1996–2008 using a triad of the regression models. The models and the proxies have been examined in previous trend studies. We adapt the MARS and bootstrap methodology to calculate the models' constants. This approach has been proposed in our previous papers (Krzyścin et al., 2005; Krzyścin, 2006; Krzyścin and Borkowski, 2008). We focus on the trend values in the period 1996–2008 as the ozone decline in the extratropical regions in the 1980s and early 1990s was firmly corroborated by many earlier studies (e.g. WMO 2007, Vyushin et al., 2007; Yang et al., 2009).
It is not clear which of the trend models is the best as each model has advantages and disadvantages. Our objective is not to search for such a model but to analyse output of all the models to obtain a more robust view on the trend variability since the mid-1990s. There are basic conceptual differences between the models (e.g. more or less freedom is selection of the trend pattern, including or omitting proxies parameterising natural variability of total ozone). EESC model calculates the trend directly linked to changes in the ODSs concentration in the stratosphere. The trends by PWL and FLEX allow to combine the ODSs changes effect with other yet unknown forcing on the ozone layer. FLEX model has not been widely used in the recent statistical trend estimates. This and our previous papers confirm its effectiveness in trend modelling. The main advantage of FLEX is the possibility to extract the trend pattern from the ozone time series. EESC and PWL models need a priori selection of the trend shape. However, FLEX model requires a priori selection of the smoothness level of the trend curve. An agreement among the statistically significant trend values by all the models provides strong support for ozone layer recovery.
All examined options of trend models give statistically significant positive trends in the period 1996–2008 for the ozone data averaged over the globe (boreal spring), and in 50°S–50°N zone (boreal summer and whole year). It constitutes there a strong support for the second stage of ozone layer recovery. Our estimates of the total ozone trend of about 0.5–1% per decade correspond to those found in the previous studies by Vyushin et al. (2007); Yang et al. (2009); and Angell and Free (2009). For other regions and seasons, the best we can say is that the steady ozone depletion as seen in the 1980s and early 1990s has been stopped, i.e. the ozone recovery is in its first stage according to the WMO (2007) definition. Evidently, a longer time series and/or new proxies resolving dynamical variability of ozone are necessary to discuss the ozone problem.
This study has been partially funded by the State of Environment in Poland under Contract No. 40/2008/F. The author would like to thank anonymous reviewers for constructive comments and suggestions which helped improve this paper.