Previous: Ext. Abst. Next: Conclusions Up: Ext. Abst.

Introduction

Artificial Neural Network (ANN) techniques have been increasingly applied in the last years to geophysical variables like solar activity [Calvo et al., 1995; Verdes et al., 2000]. They have also been used for dendroclimatic time series reconstruction [D'Odorico et al., 2000.

Different models have been developed to understand the space-time evolution of ozone just during or after the Pinatubo eruption in June 1991 at Fillipines (15° N latitude). Some of them are the one-dimensional (1D) radiative transfer model of Kinne et al. [1992], the 2D chemical-radiative-dynamical model of Brasseur and Granier [1992]. Also a linear statistical analysis have been performed by Randel and Cobb [1994]. In the present work we compare the ozone losses derived from the registered TOMS/Nimbus 7 data and the corresponding predicted ANN values in the post-eruption period, with the 2D (latitude-height) chemical-photodissociation-transport aerosol-induced perturbations model of Pitari and Vizi [1993], and with a new 2-D chemical-radiative-dynamical global model [Rosenfield et al, 1997], which includes heterogeneous chemistry not only on the surface of sulfate aerosols but also on polar stratospheric clouds.

The influence of sulfate aerosol produced by the volcanic Pinatubo eruption in the Southern Hemisphere ozone distribution has been analyzed in different works [see, for instance, Hofman et al., 1993; Deshler et al., 1994], directly from integrated or profile measurements. In the present study the band (55°- 60°)S has been considered since, besides its intrinsic interest, the city of Ushuaia in Argentina, located at this high SH latitude, is currently affected by the ozone depletion and in some moments of the year directly by the Antarctic ozone hole. Also, it is the highest latitudinal band that can be considered from TOMS ozone data, since more polar ones do not contain information all the year round.

A detailed performance comparison of different multi-dimensional models of ozone has been made in the WMO/UNEP "Ozone Assessment 1984" [WMO, 1995]. Models that include heterogeneous chemistry on aerosol surfaces plus gas-phase reactions give ozone decreases in the 1980-90 period that are in good agreement with the observed trend. Polar stratospheric clouds chemistry has been analyzed with two and three dimensional models. The WMO/UNEP "Ozone Assessment 1998" [WMO, 1999] describes the latest contributions to the understanding of the ozone behavior and the predictions of its future trend, but only as annual means or as a monthly means, and in this last case for the year 1990, before the event considered in the present work.

Figure 1. Ozone layer in the 55ºS ? 60ºS Southern Hemisphere latitudinal band, measured by TOMS, and two different ANN predictions for the "ideal" non-eruption situation. Also shown the mean annual wave.

The normal way to extract ozone trends and losses due to atmospheric perturbations from actual satellite time series is to build a comprehensive mathematical expression that includes the main variations (antropogenic trend, solar, QBO, ENSO) and to take the difference between this ideal formulation and the actual behavior [see for example Randel and Cobb, 1994, and references therein]. Interesting conclusions concerning the relative importance of each event have been obtained when the signal is rather strong. However, the final results depend on the way the series for each event are chosen. For example, the solar cycle activity variation is sometimes determined through the 10.7 cm Sun radiation emission, which is one of the best indicators of its activity, but there are others like the sunspot number or several line intensity variations (some of them in the UV) that eventually can be more reliable for analyzing the effects on ozone. Also, the QBO and ENSO are based on tropical data and its teleconnection with ozone variations at high latitudes is difficult to reproduce with precision. Another source of uncertainty is the fact that different ENSO indexes can be defined.

In the present work we follow a different approach, which consists in the use of the ANN non-linear technique for extrapolating the total ozone time series to the post-Pinatubo eruption period (June 1991-April1993), without any assumption about the particular behavior of events that could influence the geophysical variable analyzed. In Fig. 1 we show results for the ozone layer in the 55ºS ? 60ºS Southern Hemisphere latitudinal band, measured by TOMS, and two different ANN predictions for the "ideal" non-eruption situation.

Figure 2. a) Relative ozone variation determined as 100*(TOMS - ANN)/ANN for the post Pinatubo eruption period (after June 1991), considering the two different ANN results as references (ideal) situation without Pinatubo effect (solid square). The corresponding mean value is also displayed (solid square). b) Absolute difference between both ANN's given in a), with a maximum value of 2.74.

In Fig. 2a we give the mean relative difference between TOMS data and ANN predictions. Both ANN results gives essentially the same time dependence of total ozone. In order to compare with the actual loss mainly due to aerosol from Pinatubo eruption perturbation, the mean of both ANN results was determined. The maximum difference between both ANN can be used for estimating the error in the ozone loss derivation (see Fig. 2b). It must be pointed out that some contribution to the ozone loss at this high SH latitudes can also be ascribed to the eruption of the small volcano Mount Hudson, situated in the patagonian Andes mountains (46º S), in August 1991, even if its perturbation was mainly concentrated in the troposphere [Dreshler et al., 1992].

Figure 3. Relative ozone variation determined as 100*(TOMS - ANN)/ANN for the post Pinatubo eruption period (after June 1991), considering the ANN result as a reference (ideal) situation without Pinatubo effect (solid square). Error bars are the maximum value determined from the results of figure 2b. Model calculations of Pitari and Vizi (1993) (open up triangle) and of Rosenfield et al. (1997) (open down triangle).

In Fig. 3 we represent the mean predicted behavior of the ozone time series in the post-Pinatubo eruption period, in the ideal situation in which no effects of this eruption were present at the considered SH latitudinal 50°S - 60°S band. The mean values are affected by the maximum error determined in Fig. 2b. We can see that the oscillation would have continued if no strong eruption had occurred in June 1991, injecting enough sulfate aerosols and contributing significantly to the (otherwise mainly antropogenic) ozone loss [WMO, 1994 and 1999]. This behavior cannot be reproduced with a simple "mean annual wave" prediction curve, as can be easily observed from Fig. 2a.

In particular, Fig. 3 shows the percentage relative difference between TOMS data and ANN predictions, giving an almost zero ozone loss due to Pinatubo aerosols injection in the stratosphere up to nearly two months after the eruption, with a small positive contribution of about 2% in the following 3-4 months interval. A first decrease of about 4 % is apparent at the end of 1991, probably due to the vicinity of the Antarctic ozone hole. Then, another positive contribution is observed at the 1992 SH autumn, also of about 2 % on average. The SH winter and spring seasons show a strong ozone loss at this high latitudes with a maximum loss of -12 % in December, followed by a steady decrease up to the end of the analyzed period, were the Antarctic ozone hole develops. The sulfate aerosols contribute more significantly to the losses than photolysis and heating rates perturbations.

In order to compare the variations in the ozone total column derived from the ANN-TOMS data with the model calculation results for the post-Pinatubo period, the last ones has been determined for the middle of the band, i.e. 57.5° S. The calculations of Pitari and Vizi (1993) show a rapid increase starting only about two months after the eruption and producing an ozone loss with a maximum of -6%, followed by a plateau at the SH summer. The theoretical values obtained by Rosenfield et al. (1997) start with a positive variation during approximately a year. By the middle of 1992 the variation changes sign showing a strong decrease of nearly ?14% in September, with a delay of around 2.5 months. The decrease in 1993 follows closely the ANN-TOMS data.