Click for next page ( 307


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 306
Appendix E TREND ANALYSIS OF TOTAL OZONE Hans A. Panofsky Department of Meteorology Pennsylvania State University INTRODUCTION Total ozone has been measured at Arosa for almost 50 years by the Dobson spectrophotometer. The number of Dobson stations in the global observing network increased slowly at first, but more rapidly since the late 1950s. Measure- ments made with instruments developed in the USSR were included in the network system after 1958, but their reliability approached that of the Dobson measurements only since about 1972. Most of the observing stations are on continents and in the northern hemisphere. Analysis of these records by Angell and Korshover (1981) reveals considerable differences in long-term variations from region to region; but, on the average, trends appear to be mostly positive in the 1960s and near zero in the 1970s. As we will see, more recent analyses based on sophisticated statistical models suggest positive trends in the 1970s, which are, however, not significantly different from zero. Whereas it is relatively easy to estimate trends from Dobson data, it is much more difficult to ascribe such trends to specific causes. There may have been trends in variables such as dust that influence the measured ozone but not necessarily the actual ozone; there may have been real changes of ozone due to changes in circulation of the atmosphere; there may have been changes of temperature or of various trace elements that influence the ozone budget; and there may have been influences of solar varia- tion. Further, the Dobson network may not be representa- tive of global averages. As we shall see, some causes of trends in Dobson- measured ozone can be evaluated by sophisticated statistical analysis; for other causes we can only make some not very well educated guesses. 306

OCR for page 306
307 Ultraviolet satellite measurements of total ozone trend largely avoid the problem of spatial represent- ativeness (not completely, because the dark polar cap remains unobserved). However, satellite records so far are relatively short and have other deficiencies. Esti- mates of total ozone have been made by backscattered ultraviolet (B W) on the Nimbus 4 satellite beginning in April 1970. However, this instrument initially suggested global averages 3 to 4 percent less than the total Dobson averages. This difference increased further over 81 months due to instrumental drift. In addition, failure of a solar panel in June 1972 reduced the number of observations after that date, introducing a problem of spatial representativeness of the trends over the whole 81 months. The Solar Back-Scatter (SB W) instrument on Nimbus 7, operational since 1978, did not have these problems (see Hudson et al. 1982). But the period of its operation so far is too short to make reliable trend estimates. In principle, however, satellites should eventually lead to better trend measurements than Dobson instruments for two reasons: (1) there is better spatial representation, and (2) only one instrument is used, whereas the different Dobson instruments have separate idiosyncrasies and errors. Gradual instrumental drift could be estimated by comparison with a well-calibrated Dobson network. Pittock (see Hudson et al. 1982, p. 3-46) estimates that eventually satellites will be able to detect True global trends" with standard deviations of the order of 1 or ~2 percent from observations by the same satellite over 10 years. Such trends, of course, could have man-made as well as natural causes. So far, satellites, just as Dobson instruments, have not suggested any significant trend in total ozone. Most recent physical models also suggest very little ozone change due to human interference. Therefore there is, at present, no real disagreement between statisti- cians and physical modelers. The main disagreement remaining concerns the extent to which statistical analysis can be used as an early warning system for the future. For the Dobson network, there are two main problems: first, how representative is the Dobson network of the global ozone distribution? This problem has been attacked by use of satellite data. The results are somewhat controversial, as we shall see.

OCR for page 306
308 The second difficulty involves the importance of "natural" variations over long periods. This question has been attacked by studying the relatively long Arosa record. Again, the results are controversial. For satellite early warning, only the second problem is important. STATISTICAL ANALYSIS OF THE DOBSON NETWORK Three separate groups of statisticians (Bloomfield et ale 1981, Reinsel et al. 1981, St. John et al. 1981) have analyzed monthly averages at 36 Dobson stations with records over 10 years long. Their studies differ in detail but have many common features: the results also are quite similar. All of the groups originally based their analyses on the "hockey stick" or "boomerang" approach. They argued that since there was no obvious cause for the upward trend in the 1960s, the observed upward trend must be due to "red noise"--natural long-period variations. Therefore the "true" long-range trend up to 1970 was assumed to be zero. Given this zero trend, estimates for the Dobson mean trend in the 1970s were made by various statistical procedures. Such trends were generally positive (Figure E.1) but did not differ significantly from zero. 1960 1970 1980 YEAR FIGURE E.1 Typical ozone variations with time, and "hockey stick" fit (solid line).

OCR for page 306
309 Bloomfield et al. (1981) also produced separate trend analyses for the 1960s and 1970s (no hockey stick assumed), and obtained a near-zero trend for the 1970s. Further possibilities tested by Bloomfield et al. (1981) were based on assumed relationships between ozone and nuclear tests, and between ozone and solar activity. Although the trends in the 1970s derived from this analysis did change somewhat, they still are not significantly different from zero. With 36 individual trends, it has been possible to assess the uncertainty in the mean trend of the 36 stations, based on the 10- to 20-year records. Statistically, this variability from station to station was split into (1) random effects at each station; (2) variation among stations in each region (the regions are large, e.g., North America is a single region); and (3) variation among regions. However, it should be noted that these three sets of differences cannot be associated with physical causes on a one-to-one basis. For example, "random" errors are introduced in the monthly averages by missing data, by the effect of large variations with periods of the order of a week, by local air pollution episodes or clouds, or simply by observa- tional error. Variations in trends within regions are primarily due to different weather at the stations in the same region. Different weather implies real differences in ozone, and also differences in observational accuracy; e.g., clouds, which interfere with the accuracy of the ozone observa- tions. Also, rates of deterioration of instrumental parts may be different, and calibrations may have been performed at different points in the solar cycle. Weather differences are even more important in different regions; further, each region uses different secondary standards for instrument calibration. All groups agree that the standard deviation of average trend derived from these stations, due to a combination of all these factors, is about 0.6 percent per decade. However, the question of trend "bias" due to the assumption of the "hockey stick" model has been raised frequently and requires further analysis. SPATIAL BIAS In principle, it is possible that trend estimates from the Dobson network would differ systematically from

OCR for page 306
310 global trends; e.g., if ozone moved from land areas to ocean areas for an extended period, there might be an indicated ozone change but no real global trend. Three techniques have been used to estimate the bias of global trend estimated from the Dobson network. In the first technique, Hasebe (1980) has estimated total ozone values at grid points over the world from the Dobson network by a process called "optimum interpola- tion." However, since there are huge areas of no ozone stations, particularly in the southern hemisphere, the interpolated (and sometimes extrapolated) values in such areas are extremely uncertain. Hasebe computes global trends from these grid-point values. These do not differ significantly from those estimated by Angell and Korshover (1981). In any case, it is unlikely that Hasebe's techniques for estimating global trends are necessarily superior to the simpler methods used by the earlier authors. In a second technique, Moxim and Mahlman (1980) compared global trends with Dobson location trends as computed from a "simple" three-dimensional numerical global model of the atmosphere, containing ozone. They find differences between one-year trends computed from monthly averages between global and Dobson location ozone of the order of 1 percent. Finally, several groups of authors have attempted estimates of spatial bias by use of the B W satellite observations, described in the introduction. Of course, it is difficult to evaluate the accuracy of these comparisons, since the B W instrument deteriorated after several years, and reliable trends over periods longer than three years or so could not be computed. London and Ling (1981) compared global total ozone with Dobson location ozone and found only small mean differences, but large standard deviations of these averages. However, they did not compare trends. Reinsel et al. (1982) compared statistics of ozone trends all over the world with data for April 1970 to May 1975 with those of ozone trends in small areas surrounding a select group of 36 Dobson stations. They found no statistically significant differences in this one small sample covering only a few years. The authors concluded from this result that there are no differences between global trends and trends derived from Dobson station data. Meteorologists are doubtful about this conclusion and suggest standard deviations of global trends due to Dobson location bias could well be of the order of 1 percent per

OCR for page 306
311 decade. This order of magnitude is also suggested by satellite records analyzed by A.J. Miller (see Hudson et al. 1982), and is somewhat smaller than what would have been expected from Moxim and Mahlman's model study. In fact, the results of significantly from an assumption of a bias of decadal average Dobson trend of 1 percent. Hence, meteorologists generally suggest a spatial bias in the Dobson network, of the order of 1 percent per decade, but realize that this is an extemely uncertain number. . ~ . . . ~ _ _ ~ ~ _ ~ . , Reinsel et al. do not differ LOW-FREQUENCY VARIATIONS Bishop and Hill (1981) used a partially inhomogeneous Arosa record to estimate the trend uncertainty due to low-frequency variations (not directly analyzable with the record of the Dobson network) from the variation among decadal trends. Their result was an uncertainty of the order of 0.8 percent per decade. The Arosa record used by Bishop and Hill is remarkable for the absence of fluctuations with periods of the order of a century, periods that are often quite apparent in temperature records. In fact, the homogenized Arosa ozone record for the period 1932-1980 (Dutsch, personal communication to J. London, University of Colorado, 1981) shows noticeable long-period variations (Figure E.2). The figure also shows the sensitivity of 10-year trends to the starting date of each decade. For example, the 360 350 c c o o 340 at o o 330 ~ 320 I'. 41 1 ~ 11 1 , \/ t1 I V `! 11 I] 1 1 1~ _ 1 1 1 1 1 1 1926 1930 1940 1950 1960 1970 1980 YEAR FIGURE E.2 Homogenized values for total ozone at Arosa (Hudson et al. 1982).

OCR for page 306
312 trend for 1936-1945 is almost zero. for 1940-1949 is -33 Dobson units. In contrast, that Hence, the estimate of 0.8 percent for long-period variability is quite uncertain. Further, this estimate is based on the record from only one station. A further indication of the magnitude of uncertainties due to long-period fluctuations is the sensitivity of the trends in the 1970s to the statistical model used. Thus, the "hockey stick" approach and the hypotheses of separate unknown trends in the 1960s and 1970s produce trends in the 1970s differing by 1.5 percent. E.L. Scott (University of California, Berkeley, personal communica- tion, 1981) also has criticized the dependence of the global trend estimates on the particular statistical model chosen. For these reasons, we suggest that the uncertainty of global decadal ozone trends due to long- period natural variations with unknown causes is at least of the order of 1 percent; but we consider this estimate also as very uncertain. SUMMARY AND RECOMMENDATIONS Because of the various controversies, it is not possible to arrive at a definite understanding of the uncertainties of global decadal ozone trends derived from Dobson stations or satellites. The statisticians using the "hockey stick" models would estimate the standard devia- tions of the decadal trends derived from Dobson stations to be of the order of 1 percent. Most meteorologists and at least one statistician would prefer standard deviations of the order of 2 percent or larger. Once long-lived satellites with good reliability characteristics are available, the standard deviations may perhaps be cut considerably. As we consider periods longer than 10 years, the trends are better determined but they have to be extrapolated to a longer period. These two factors almost cancel; the uncertainty of trend may decrease only slightly for longer periods. Therefore, meteorologists would consider trend analysis an unreliable early warning system for man-made ozone changes; an observed average trend of 0 percent at the Dobson stations, over, say 20 years, could be produced by a man-made decrease of 4 percent or larger, compensated by other factors, with a probability of 5 percent or so. If remedial action is taken only after a "significant" decrease of ozone, existing fluorocarbons could continue

OCR for page 306
313 to decrease the ozone further to intolerable levels due to their long lifetimes. Some statisticians would consider a zero trend as an indication that the trend due to fluorocarbons must have been much smaller than 4 percent. Our best hope for using total ozone trends as early-warning systems rests on improved and well- calibrated records of total ozone measured from long-lived satellites. REFERENCES Angell, J.K. and J. Korshover (1981) Update of ozone variations through 1979. Pages 393-396, Proceedings of the Quadrennial International Ozone Symposium, August 4-9, 1980. Boulder, Colo.: National Center for Atmospheric Research. Bishop, L. and W.J. Hill (1981) Analyzing Stratospheric Ozone for the Natural and Man-made Trend Variability. Pages 304-305, Summaries of Conference Presentations. (Available from the Society for Industrial and Applied Mathematics, Philadelphia, Pa.) Bloomfield, P., M.L. Thompson, G.S. Watson, and S. Zeger (1981) Frequency Domain Estimation of Trends in Atmospheric Ozone. Technical Report No. 182, Department of Statistics, Princeton University. (Submitted for publication to Journal of Geophysical Research) Hasebe, Fumio (1980) A global analysis of the fluctuations of total ozone II nonstationary annual oscillation, quasi-biennial oscillation, and long-term variations in total ozone. Journal of the Meteorological Society of Japan 58:104-117. Hudson, R.D., et al., eds. (1982) The Stratosphere 1981: Theory and Measurement. Geneva: World Meteorological Organization. (Available from National Aeronautics and Space Administration, Code 963, Greenbelt, Md. 20771) London, J. and X. Ling (1981) The geographic bias in determining average variations of total ozone from ground-based observations. Pages 337-339, Proceedings of the Quadrennial International Ozone Symposium, August 4-9, 1980. Boulder, Colo.: National Center for Atmospheric Research. Moxim, W.J. and J.D. Mahlman (1980) Evaluation of various total ozone sampling networks using the GFDL 3-D tracer model. Journal of Geophysical Research 85 (C8):4527-4539.

OCR for page 306
314 Reinsel, G., G.C. Tiao, M.N. Wang, R. Lewis, and D. Nytchka (1981) Statistical analysis of stratospheric ozone data for detection of trend. Atmospheric Environment 15:1569-1578. Reinsel, G., G.C. Tiao, and R. Lewis (1982) A statistical analysis of total ozone data from the Nimbus B W satellite experiment. Journal of Atmospheric Sciences 39. St. John, D.S., S.P. Bailey, W.H. Fellner, J.M. Manor, and R.D. Snee (1981) Time series analysis for trends in total ozone measurements. Journal of Geophysical Research 86:7299-7311.