Trends of Air Pollution in the Fichtelgebirge ... - Semantic Scholar
Trends of Air Pollution in the Fichtelgebirge mountains, NE Bavaria Otto Klemm and Holger Lange Universität Bayreuth Bayreuther Institut für Terrestrische Ökosystemforschung (BITÖK) D-95440 Bayreuth Germany manuscript, submitted for publication in Environmental Science and Pollution Research in revised form, 19. February 1999 key words: air pollution; atmospheric chemistry; autocorrelation; complexity analysis; environmental pollution; Hurst statistics; nitrogen oxides; ozone; deposition; recurrence quantification; sulfur dioxide; time series; trends
Abstract. We analyzed 13 years of hourly measurements of SO 2, NOx, and O3, at forest ecosystem research sites in SE Germany. A quasi continuous data record was obtained by combining data sets from two locations. Before interpreting trends in the combined data set, we analyzed if the change of location introduced a systematic bias. We employed autocorrelation functions, Hurst statistics, complexity analysis, and recurrence quantification and found that the partial data set exhibited no indication of the presence of any bias. For SO2, we also compared the data from the forest sites with data obtained in nearby cities and also found no indications for any systematic effects. Applying nonparametric trend statistics we found a significant decrease of the SO 2. Most of the observed decrease is due to the reductions of SO 2 emissions in eastern Germany, but reductions in western Germany and the Czech Republic also played important roles. For O3, we observed a significant increase, the causes of which are unclear from our data alone. For NOx, no trend was identified.
1. INTRODUCTION The levels of air pollution have undergone significant changes in central Europe within the last few years. Particularly, emission reductions of sulfur gases have led to lower levels of sulfur dioxide (SO2) at various stations in Germany and other places. For western Europe, these reductions have mainly resulted from regulations that went into effect in the mid to late 1980’s. In eastern Germany, the breakdown of the economy, that came with the German unification in 1990, also resulted in a strong decrease of air pollutant emissions. For various countries in eastern Europe, less is known about the time pattern of emissions, although a certain decrease of the emissions is likely to also have occurred in these areas. In this contribution, we analyze to what extent the emissions reductions are detectable in routine measurements of air quality in Central Europe. Our area of interest is the Fichtelgebirge, NE Bavaria, a mountainous range in the center of Europe, reaching peak altitudes of just over 1000 m a.s.l. This area is strongly affected by forest decline phenomena (SCHULZE et al., 1989). It is of great interest in our ecosystem research studies to know whether a significant change in air quality has occurred over the years. 2. DATA BASE Most of the routine measurements of air quality in Germany are conducted in the centers of cities. These data are used to decide if the levels of air pollution are likely to affect the health of the human population. We are, however, more interested to know whether the vegetation in rural areas is likely to be affected by air pollution or not. As part of a program to study the forest decline phenomena in Northern Bavaria, one station has been set up in 1985 at a remote, forested site close to the village of Warmensteinach in the "Fichtelgebirge" at an altitude of 575 m a.s.l. After operating at that site until 1993, the station has been moved northward to the research site "Waldstein", at 765 m a.s.l. at a distance of 18 km from the earlier site in Warmensteinach. The move led to a gap in the data series of 8 months duration. In combination, a 13 year time series of air quality measurements is available for the Fichtelgebirge region. However, there are some limitations concerning the interpretability of the data set. First, it is not clear whether the move led to any systematic bias between the two data subsets and secondly, the locations
of both stations are in small forest clearings, so that the respective micrometeorological conditions are heavily modulated by the local topography and vegetation. Within our data analysis, we want to estimate where air masses with high or low trace gas concentrations originate. For example, we want to quantify if, in a given time period, air masses of very high SO2 concentrations come from Eastern Germany, located to the North of our research station, from the Czech Republic in the East, or from western Europe, located in the West and South. However, it is not feasible to calculate a large number of backward trajectories on a daily or even more frequent basis. We made use of the meteorology data that were acquired by the Bavarian State Ministry for State Development and Environmental Affairs (StMLU) and by the German Weather Service at their stations outside the cities near the Fichtelgebirge mountains (Bayreuth, Kulmbach, Naila, Hof, Selb, Arzberg). We define a mean wind vector of the Fichtelgebirge, estimating the regional flow field to the best possible extent, by averaging all available wind vectors from the city stations for every hour of measurement. Although the physical meaning of a mean vector is limited because different stations exhibit different wind directions at a given time, we suggest that this vector is a much better representation of the regional flow pattern than the winds measured at the forest stations. Hereafter, the hourly averaged spatial mean of the wind vector will be used to interpret any measured trace gas concentrations with respect to the associated wind directions. Further, we created a synthetic data set for SO 2 that represents the SO2 concentrations in the cities at the upwind side of the Fichtelgebirge mountains. For situations with winds from NW to NE (mean wind direction between 315 and 45 degrees), the mean SO 2 concentrations of the cities of Naila and Hof were used. For easterly winds (45 through 135 degrees), the averages of the stations Selb and Arzberg were used, and for the remaining data (wind from 135 through 315 degrees), we used the data from Bayreuth and Kulmbach. In our region, high SO2 concentrations occur in episodes that last between a few hours and several days. It appears from a visual inspection of the original data sets that episodes of high SO2 occur simultaneously at the forest sites and in the cities upwind of the Fichtelgebirge, respectively. Further, it appears that the two forest sites reflect the pattern of the SO2 concentrations in the Fichtelgebirge in a similar manner. They have indistinguishable Pearson correlation coefficients with the synthetic
data set for the respective periods (r2=0.50 and r2=0.46 for Warmensteinach and Waldstein, respectively). The SO2 data from the forest sites (see Fig. 1, solid line, for a frequency distribution) cover a wide range of up to 250 ppb. More than half of the data are at the lower edge of the scale between 0 and 4 ppb. This leads to an extremely skewed distribution (nondimensional skewness = 4.907, which is about by a factor of 400 larger than the skewness of a Gaussian distribution of the same sample size) which can not be appropriately described with a parametric distribution. Further complications arise from the fact that the low mixing ratios are in many cases close to or below the detection limit of the employed analytical instruments. For the analysis of NOx and O3, we rely exclusively on the data from the forest sites in Warmensteinach and at the Waldstein because the city data of NOx and O3 are governed by the local road traffic, which injects NO into the atmosphere that reacts with O 3 within a few minutes. NOx measurements were conducted at Warmensteinach only between September 1988 and December 1990. At the Waldstein site, NOx data are available for the entire period between 1994 and 1997. The O3 data of the forest sites cover the entire period 1985 - 1997 with some large (1 to 2 months duration each) gaps in the data records between 1985 and 1993. At the Waldstein station (1994 through 1997), the O 3 data record is almost complete. All trace gas data are given in units ppb, which means a volume (or molar) mixing ratio of 10-9. Note that other publications and data bases sometimes use concentrations units such as µg ⋅ m-3. In many cases, the concentrations refer to a volume of air under standard conditions (1013 hPa, 298 K). For these conditions, a mixing ratio of 1 ppb equals a concentration of 2.6 µg ⋅ m-3 in the case of SO2 and 2 µg ⋅ m-3 in the case of O3. 3. TIME SERIES INVESTIGATIONS A variety of investigation tools yield information about periodicities, persistence, instationarities, and trends in the data sets. The characterization of the data sets for SO2, NOx, and O3, is guided by two questions: First, are there indications for a significant bias induced by the transfer of the forest measurement station from Warmensteinach to the Waldstein site in 1993/1994? And secondly, are there any observable long-term
trends? For an evaluation of the first question, we calculate the autocorrelation functions (section 3.1), Hurst statistics (3.2), two complexity measures (3.3), and recurrence quantification parameters (3.4) for subsets from Warmensteinach and Waldstein, respectively. For a clarification of the second question, we also use the results from sections 3.1 through 3.4 for a characterization of the time series with respect to the presence of periodicities and long-term as well as short time behavior, and, in addition, we perform a trend analysis in section 3.5. Because of the gap within the data sets that resulted from the move of the station in 1993/1994, it is not meaningful to calculate any statistical measures that require equidistant data for the entire period of investigation. We analyze two representative data subsets of approximately equal length from Warmensteinach (1985 - 1987) and from the Waldstein (1994 - 1997), respectively. These measurement periods are also not completely gap-free. Nevertheless, we tried several different interpolation schemes to fill the gaps in the data sets and found that our analyses were not significantly affected. Therefore, we ignored the data gaps in the following analyses.
3.1 AUROCORRELATION FUNCTIONS One of the standard tools for linear time series analysis is the autocorrelation function (ACF) r (k ) of a data set x (ti ) obtained from the autocovariance
φ (k ) =
1 N −k ∑ ( x(ti +k ) − x )( x(ti ) − x ) N − 1 i =1
and r ( k ) = φ (k ) / φ (0) = φ ( k ) / σ x2
where k is the time lag considered, N the length of the series, x its overall arithmetic mean, and σ x2 its variance.
The ACF quantifies linear dependencies between values within a data set; all r (k ) lie between –1 and +1, and values significantly above zero denote correlation at the given lag. A negative r (k ) indicates significant anticorrelation. Existing periodicities are exhibited by regular peaks in r (k ) ; their heights may be used to estimate their importance relative to other (e.g. non-periodic) correlations. A finite memory of the series leads to a gradual decline of the envelope of the ACF. The autocorrelation functions for our series are shown in Figure 2. There is a well pronounced seasonality for O3 at the Waldstein site. This phenomenon is mainly due to the fact that O3 is predominantly produced during summer due to high NO 2 photolysis frequencies. However, the amplitude of the autocorrelation function becomes less pronounced for longer time lags (>1 a), and after 3 years, the autocorrelation coefficients reach only values of around 0.1. This is an indication that the memory of the O 3 signal is finite, i.e, an autocorrelation length is calculable, and that the periodicity is not extremely regular. It may also point to a trend in the data set. On the other hand, the plausible finite size effects at large time lags usually lead to large fluctuations in the ACF that are not observed here. For the Warmensteinach site, the seasonality is also indicated; however, in this case, the measurement period is too short to lead to definite conclusions. The seasonality for NOx is also clearly established. As the emissions of NOx are typically not higher in the winter months than during the summer, this result is, at first glance, surprising. We assume that the origin of the seasonal pattern of NOx lies in the different meteorological conditions during winter rather than in different NO x emission conditions: In winter, the near-surface atmospheric boundary layer is generally less turbulent and less deep. Therefore, the NO x emssions from surface sources are less well mixed within the lowest layer of the atmosphere and therefore establish higher concentrations of NOx at a site on the ground. The autocorrelation of SO2 also reaches significant values on an annual cycle, as would be expected from the observation that episodes of high SO2 are more frequent and more pronounced during winter than during summer. However, the autocorrelation of SO 2 phases out rather quickly (within about two years), pointing to a much more pronounced finite memory range than in the O3 case, and also to the presence of a trend.
3.2 HURST ANALYSIS Further insight into the extension and importance of memory effects and the long-term structure of our atmospheric time series is gained by calculation of the Hurst exponent H. This measure quantifies the extension of periods with systematic deviations from the overall mean, which is also called the persistence of the time series. If the value range of a time series scales with the length of the time series, the Hurst effect is present. The best fit value for H found typically in experimental data sets is 0.5 ≤ H ≤ 1.0; the lower limit corresponds to ordinary Brownian motion (low persistence), the upper one to very high persistence; values in between correspond to the so-called fractional Gaussian noise (RODRÍGUEZ-ITURBE and RINALDO, 1997). There are different techniques to estimate H. Our calculation was performed with the rescaled range or R/S statistics (MONTANARI et al. 1997). In all our analyzed data sets, persistence is clearly present (H > 0.85). The H values for O3 from Warmensteinach (0.92 ± 0.03) is remarkably similar to the one from the Waldstein (0.93 ± 0.01). Also for SO2, the H values from the two sites are very similar (0.87 ± 0.03 for Warmensteinach and 0.86 ± 0.02 for Waldstein, respectively). On the other hand, for a single research site, the difference between the two gases is highly significant (e.g., 0.93 for O3 vs. 0.86 for SO2 at the Waldstein site). We conclude that the Hurst exponent is a good characterization of the respective gases and that the characteristics are almost identical for a single gas (e.g., O3) in the two data subsets. There is no indication of a systematic bias in the two data sets. The particularly high H exponents of O3 probably originate in the fact that high O3 typically occurs during photochemical episodes in summer that typically last several days. For SO2 and NOx, the Hurst exponents are also quite high (0.88 ± 0.03 for NOx at the Waldstein site), again emphasizing the episodic nature of events with very high concentrations. 3.3 COMPLEXITY MEASURES Short term structures of the time series are investigated with the aid of complexity measures. This serves to elucidate the relationship between randomness of a data set (in many cases correlated with noise level) and its "true" complexity (BAR-YAM 1997). The use of this concept refers to the difficulty to describe the temporal structure in a compact
way. When quantified, complexity should be low for nearly constant or slowly-varying signals, and should also be low for nearly completely random sequences as the statistical description of a white–noise process (a completely unstructured process with no correlations among consecutive values, e.g., an evenly distributed random distribution) is easy. Calculation of these measures provides information about the optimal temporal resolution for measurements of a variable (LANGE et al. 1998, LANGE 1999). The interplay between driven and independent parts, or between underlying processes of different time scales, determines the quantification of randomness and complexity. Technically, complexity measures are calculated on the basis of a discretized version of the original series, the so called symbol sequences (WACKERBAUER et al. 1994). Small groups of consecutive values within these sequences are called words, the length L of which has to be fixed (we have chosen L = 4 for all investigations here). The method quantify the short-term correlations within the range prescribed by the word length L in an average manner. Complexity is quantified as fluctuation complexity (BATES AND SHEPHARD, 1993), which vanishes both for constant as well as random data and has a maximum in–between constant and random data. Our preferred randomness measure is the Mean Information Gain (MIG, W ACKERBAUER et al. 1994). The MIG is minimal for constant and maximal for random sequences. We present the calculations in complexity diagrams where MIG is the abscissa and the fluctuation complexity the ordinate. The curves in Figure 3 were obtained by varying the temporal resolution by aggregation (1 – 48 h). As expected, the randomness measure increases as more and more short–term correlations are averaged out, whereas the fluctuation complexity first increases and then decreases. The difference between SO2 and O3 at high aggregation level is obvious, as is the remarkable similarity between the two subsets. This indicates that the small–scale structure of the data sets is not influenced by the move of the research site from Warmensteinach to Waldstein. The distance of the randomness/complexity values from the theoretical limit (obtained for a Bernoulli sequence) is larger than for typical hydrological variables (L ANGE 1999). At the same level of randomness, less complexity is found in our data sets for atmospheric trace gases than for water signals (i.e., runoff from catchments). The fact that the O3
signal becomes insensitive with respect to randomness for sufficiently high aggregation level stems from the daily cycle of photolysis; for a strictly periodic signal with period P, the MIG is constant once the words span more than the period ( L ≥ P ). In our case, with L = 4 and hourly resolution, this is the case for aggregation level 24 / 4 = 6, confirmed for both measurement stations. For SO 2, no daily cycle is visible. Note that the annual cycle is far from being detectable by this method as the maximal word length is limited by the need to obtain saturated word frequency statistics. Maximal complexity values are obtained for both gases at an aggregation width of 2–3 hours. 3.4 RECURRENCE QUANTIFICATION We used the technique of recurrence plots (ECKMANN et al. 1987) and their quantification (ZBILUT et al. 1998) to estimate the degree of determinism (DOD) in the respective time series. Recurrence plots are a visualization of recurrent patterns at arbitrary locations within a time series. They are constructed from a time series of length N by firstly building embedding vectors (d consecutive values put together), and secondly calculating vector distances between them for each pair. In this way, a matrix is built and matrix elements below a predefined threshold distance r are considered as recurrent points. Deterministic (sub-)structures in the recurre nce matrix are revealed by short line segments parallel to the main diagonal. Defining a minimal line length to avoid counting accidental lines, one calculates the ratio of recurrent points which are members of lines and the total number of recurrent points to obtain the DOD. As the DOD value clearly depends on the threshold value and minimal line length chosen, one should not interpret the absolute values quantitatively. However, if we choose the same parameter set for O 3 and SO2 and consider the DOD values of consecutive (overlapping) windows of fixed length, relative degrees lead to meaningful conclusions. Figure 4 clearly demonstrates that the DOD characterizes the variable rather than the measurement period. For SO 2, the DOD is generally high (low noise level). For the Warmensteinach period, it seems to be stationary and covaries with the percentage of recurrent points, indicating the absence of structural instationarities. For the Waldstein data set, there are two remarkable exceptions to this covariation of unknown origin, and a decreasing DOD trend towards the end of the period. This may be due to the fact that
the declining SO2 concentrations also reduce the distance to the noise floor (measurement resolution). The O3 data are similar for the two stations, but drastically different from the SO 2 signal: the O3 DODs are quite low, the signal is thus substantially noise-corrupted, which is plausible from visual inspection of the time series, and a seasonal cycle is present, with higher values for DOD when O3 concentrations are low. 3.5 TREND ANALYSIS Trend statistics was calculated using percentiles of the measured mixing ratios for every year between 1985 and 1997 and applying the non-parametric Mann-Kendall test for the presence of trends (HOLLANDER and WOLFE, 1973; GILBERT, 1987) and Sen’s method (GILBERT, 1987) for determining the slopes of the trends. The results are summarized in Figure 5. We found decreasing trends of the SO 2 mixing ratios at the forest sites. All trends are highly significant (α ≤ 0.05) and we note that the highest percentiles decreased more steeply and with higher significance than the lower percentiles. For the forest sites, the decrease of the SO2 mixing ratios is further visualized in Figure 1: For high mixing ratios (> 13 ppb, or > 1.1 on the logarithmic abscissa), the relative frequency line of the Warmensteinach data set is well above the corresponding line of the entire data set and even further above the Waldstein line, indicating that higher SO 2 mixing ratios were much more frequent in the earlier data. In the cities, the SO2 were higher (about by a factor of 2, Fig. 5), but also decrease more rapidly within the 13 year period. The cities/forest ratios decreases over the years: after exhibiting ratios between 2 and 3 in the mid–1980's, the ratios seem to level out close to unity in the mid–1990's (0.996 ≤ ratio ≤ 1.052 in the year 1997). Figure 6 shows the SO2 wind roses for the earliest years of our measurements and for the latest years at the forest sites. In Warmensteinach (1985 - 1987), the highest concentrations are typically found when the winds were from the North and Northeast. 95% percentiles around 100 ppb were found for wind direction averages of 0°, 30°, and 60°, respectively. For the later data from the Waldstein site, the most important feature seems to be the strong decrease of the SO2 mixing ratios as compared to the earlier data, as it would be expected from the results as presented in Figure 5. It becomes
additionally clear from Figure 6 that the SO 2 mixing ratios decreased for all wind directions. The wind direction with the highest SO2 mixing ratios has turned clockwise in the later data sets compared with the earlier ones: at the Waldstein site, the SO2 maxima were clearly associated with winds from 60°, while in Warmensteinach the maximum covered a broader angle between 0° and 60°, with the maximum at 30°. This result indicates that the decrease of the SO2 was stronger for northerly winds than for northeasterly and easterly winds. Statistical analysis confirms this observation of a much more significant (α ≤ 0.002) and a much steeper (-3.26, -2.73, -2.33 ppb a-1 for the 95%, 90%, and 80% percentiles, respectively) decrease for SO 2 associated with northerly winds than for easterly winds (0.082 ≤ α ≤ 0.092; slope -1.54, -1.06, and -0.68 ppb a-1, respectively). Here we observe a direct effect of the decrease of the SO 2 emissions between the mid 1980's and the mid 1990's which was much more efficient in eastern Germany, located to the North of our site, than in the Czech Republic to the East (Umweltbundesamt, 1997; EMEP, 1997). The O3 mixing ratios exhibit a highly significant increasing trend at our forest sites (Fig. 5). The high percentiles of the O3 distribution increase more intensively than the low percentiles (e.g., medians). This indicates that it is the frequency and/or intensity of episodes of high O3 that dominates the general increase of O3 at our sites. For the nitrogen oxides, our data base is much weaker, and no significant trends could be found. 4. CONCLUSIONS We analyzed a 13-year time series of hourly SO 2, NOx, and O3 data from forest sites in the Fichtelgebirge mountains, NE Bavaria, Germany. We applied extensive statistical tests (sections 3.1 - 3.4) to find indications of whether the move of the forest measurement site to a different location in 1993/1994 introduced any systematic bias into our data sets. All tests, operating independently from each other and referring to different characteristic qualitites of the data sets, indicated that there are no systematic differences between the data sets from the two sites. We conclude that it is appropriate to construct a combined 13-year time series. We obtain further support for the applicability of this approach by comparing the combined data set from the forest sites with a synthetic SO2 data set representing the urban stations at the respective upwind side of the Fichtelgebirge mountains.
Most of the time, the cities show higher SO2 mixing ratios than the forest sites. City stations are, in terms of atmospheric transport time, always closer to SO 2 emission sources (large industry and power plants but also domestic lignite burning and small industry) than the forest sites. Any SO2 being advected from the upwind side of the Fichtelgebirge mountains to our forest sites is subject to dilution, deposition to the surface, and to gas and liquid phase oxidation. These processes are very likely to result in a decrease of the SO2 concentration during their transport from the cities to the forest sites. However, the differences in SO2 concentrations in cities as compared to the forest site decreased strongly between 1985 and 1997. This decrease probably reflects the fact that the reduction of sulfur emissions in cities was more effective than for the average of all emissions. The SO2 shows a particularly strong decrease in both the forest and city data sets between 1987 and 1988 due to the enforcement of air pollution control strategies in West Germany, while the breakdown of the East German enonomy after opening of the German border and resulting emission reductions between 1990 and 1991 are not reflected in our data sets. Overall, the emissions of SO 2 were reduced by about or 61% between 1985 and 1994 (Umweltbundesamt, 1997). The agreement of this decrease with the decrease of the 95%, 90%, and 80% percentiles of SO2 in the forest data set (61%, 64%, and 36%, respectively) is convincing. We conclude that the reduction of the SO2 emissions lead to a one-to-one response of the high (95% and 90% percentiles) SO2 mixing ratios at our forest sites. Nitrogen is emitted into the atmosphere mainly as NO and NH3. Atmospheric transport, reactivities, and deposition mechanisms of these species and their products (NO 2, HNO3, particulate NO3-, particulate NH4+, and others) are very different from each other. Our data base is too weak to draw quantitative conclusions, but it should be stated that the emission reductions of 29% in Germany between 1985 and 1994 are not reflected in the mixing ratios at our forest site. Also, no preference of higher mixing ratios has been found according to certain wind directions (such as for SO 2, Fig. 6). The results are in general agreement with studies from other sites in Europe, where reductions of atmospheric deposition of sulfur compounds was found but not for nitrogen compounds (AVILA, 1996).
Our O3 data show a distinct increase over the 13 year period of observation. The 95% percentile increased by almost 2 ppb ⋅ a-1. The atmospheric processes governing the production and transport of O 3 are highly complex and nonlinear and it is therefore not clear from our analysis what are the underlying causes for this increase in O 3. Any threshold or limit values of the German and European Union authorities for SO 2 and NOx (e.g., EU threshold of 100 µg ⋅ m-3 SO2 as 24-hrs-mean) were not even closely approached at any time at our forest site. For O 3, however, the EU threshold for the protection of vegetation (200 ppb one-hour-mean) was frequently exceeded. Ozone also exceeds the critical O3 level for forest protection (exposure over the threshold value of 40 ppb during the summer months and during the daylight hours, AOT40, Fuhrer et al., 1997) of 10 ppm ⋅ h since 1991, with values between 17.3 and 21.5.
Acknowledgements: This work was supported by the German Bundesministerium für Bildung, Wissenschaft, Forschung und Technologie (BMBF) through grants Nos. PT BEO 51-0339476B+C. We are indebted to the Bavarian Landesamt für Umweltschutz and the Deutscher Wetterdienst for supplying us with data from their monitoring networks. Thanks to T. Ferdelman for language-editing of the manuscript.
REFERENCES AVILA, A. (1996): Time trends in the precipitation chemistry at a mountain site in northeastern Spain for the period 1983–1994. Atmos. Environ. 30, 1363-1373. BATES, J.E. and SHEPHARD, H.K. (1993): Measuring complexity sing information fluctuations. Physics Letters A 172, 416-425. BAR-YAM, Y. (1997): Dynamics of Complex Systems. Addison-Wesley, Reading. ECKMANN, J.P., KAMPHORST, S.O. and RUELLE, D. (1987): Recurrence plots of dynamical systems. Europhysics Letters 4, 973-979. EMEP MSC-W Report 1/97 (1997): Emissions, dispersion and trends of acidifying and eutrophying agents. Norw. Met. Inst., PO-Box 43-Blindern, 0313 Oslo 3, Norway. GILBERT, R. 1987. Statistical Methods for Environmental Pollution Monitoring. Van Nostrand Reinhold. Melbourne. FUHRER, J., SKÄRBY, L. and ASHMORE, M.R. (1997) Critical Levels for ozone effects on vegetation in Europe. Environ. Pollut. 97, 91-106. HOLLANDER, M., and WOLFE, D.A. 1973. Nonparametric Statistical Methods. John Wiley & Sons. New York. LANGE, H., NEWIG, J. and WOLF, F. (1998): Comparison of complexity measures for time series from ecosystem research. Bayreuther Forum Ökologie 52, 99-116. LANGE, H. (1999): Are Ecosystems Dynamical Systems? Journal of Computing Anticipatory Systems (in press). MONTANARI, A., ROSSO, R., and TAQQU, M.S. (1997): Fractionally differenced ARIMA models applied to hydrologic time series: Identification, estimation, and simulation. Water Resources Research 33, 1035-1044. RODRÍGUEZ-ITURBE, I. and RINALDO, A. (1997): Fractal River Basins. Cambridge University Press. SCHULZE, E.-D., LANGE, O.L., and OREN, R. (Eds.) 1989. Forest Decline and Air Pollution. Ecological Studies 77, 475 pp. UMWELTBUNDESAMT, (Ed.), 1997. Daten zur Umwelt. Erich Schmidt Verlag. Berlin. WACKERBAUER, R., WITT, A., ATMANSPACHER, H., KURTHS, J., and SCHEINGRABER, H. (1994): A Comparative Classification of Complexity Measures. Chaos, Solitons & Fractals 4, 133 - 173. ZBILUT, J.P., GIULIANI, A., and WEBBER Jr., C.L. (1998): Recurrence quantification analysis and principal components in the detection of short complex signals. Physics Letters A 237, 131-135.
60 Warmensteinach 1985 - 1987 all forest data 1985 - 1997 Waldstein 1994 - 1997
d (relative frequency / %) / d (log SO2 mixing ratio / ppb)
50
40
30
20
10
0 0.6
1.0
1.5
2.0
log (SO2 mixing ratio / ppb)
Fig. 1: Frequency distributions of the hourly SO 2 data (all data and partial data sets) from the forest sites. The lower end of the distributions (SO 2 mixing ratio < 4 ppb, or log(SO2) < 0.6) are not shown because they are influenced by the detection limits of the employed analytical instrumentation.
time lag / a 0
1
2
3
0.6
Waldstein 1994 - 1997 O3
0.5 0.4 0.3
SO2
NOx
0.2 0.1
autocorrelation coefficient r
0.0 -0.1 -0.2 -0.3
0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 -0.4
Warmensteinach 1985 - 1987
O3
0
SO2
1
2
time lag / a
Fig. 2: Autocorrelation functions (ACF) of trace gas data at the two forest sites in the Fichtelgebirge mountains. Any data points above the dashed line indicate significant positive autocorrelation for the respective time lag. NO x was not measured in Warmensteinach between 1985 and 1987.
Mean Information Gain 0.2 2.0
0.3
0.4
0.5
0.6
0.7
SO2
Warmensteinach Fluctuation Complexity
1.8
O3
1.6 1.4 1.2 1.0
1.8 Fluctuation Complexity
0.8
Waldstein
SO2 O3
1.6 1.4 1.2 1.0 0.8 0.2
0.3
0.4
0.5
0.6
0.7
0.8
Mean Information Gain
Fig. 3: Fluctuation complexity (FC) versus mean information gain (MIG) of the SO2 and O3 from Warmensteinach and the Waldstein site. The solid lines describe the theoretical upper limit of the FC.
100 80
1985
1986
1995
1987
SO2
SO2
O3
O3
1996
1997
1996
1997
60 40 20 100 80 60 40 20 0
1985
1986
1987
Warmensteinach
% Recurrence,
1995
Waldstein
% Determinism
Fig. 4: Percent recurrence and percent determinism of the SO 2 and O3 data from Warmensteinach and the Waldstein site, respectively. The x-axes are the end times of the time windows (window lengths: 83 days).
100
1985
1987
80
1989
1991
1993
1995
1997
SO2 forest sites / ppb
-1
-1.84 ppb a ; α = 0.015 -1 -1.10 ppb a ; α = 0.021 -1 -0.50 ppb a ; α = 0.025
60 40 20 0 100
-1
-4.71 ppb a ; α < 0.001 SO2 cities / ppb -1 -2.98 ppb a ; α < 0.001 -1 -2.07 ppb a ; α < 0.001
80 60 40 20 0 100 80
O3 forest sites / ppb +1.90 ppb a-1-1; α = 0.003
+1.70 ppb a ; α = 0.015
60 40 20
-1
0 1985
1987
1989
+1.40 ppb a ; α = 0.021 -1 +0.70 ppb a ; α = 0.034 1991 1993 1995 1997 year
95% percentile 80% percentile
90% percentile 50% percentile (median)
Fig. 5: Annual percentiles of the SO2 and O3 mixing ratios between 1985 and 1997. Significant trends and the respective significance levels (α) of the various time series are given.
Fig. 6: Wind roses of the SO2 mixing ratios (ppb) for the forest sites. Bold diagonals indicate a sectorization of the wind rose according to estimated regional origin of the respective air masses.
Abstract. We analyzed 13 years of hourly measurements of SO 2, NOx, and O3, at forest ecosystem research sites in SE Germany. A quasi continuous data record was obtained by combining data sets from two locations. Before interpreting trends in the combined data set, we analyzed if the change of location introduced a systematic bias. We employed autocorrelation functions, Hurst statistics, complexity analysis, and recurrence quantification and found that the partial data set exhibited no indication of the presence of any bias. For SO2, we also compared the data from the forest sites with data obtained in nearby cities and also found no indications for any systematic effects. Applying nonparametric trend statistics we found a significant decrease of the SO 2. Most of the observed decrease is due to the reductions of SO 2 emissions in eastern Germany, but reductions in western Germany and the Czech Republic also played important roles. For O3, we observed a significant increase, the causes of which are unclear from our data alone. For NOx, no trend was identified.
1. INTRODUCTION The levels of air pollution have undergone significant changes in central Europe within the last few years. Particularly, emission reductions of sulfur gases have led to lower levels of sulfur dioxide (SO2) at various stations in Germany and other places. For western Europe, these reductions have mainly resulted from regulations that went into effect in the mid to late 1980’s. In eastern Germany, the breakdown of the economy, that came with the German unification in 1990, also resulted in a strong decrease of air pollutant emissions. For various countries in eastern Europe, less is known about the time pattern of emissions, although a certain decrease of the emissions is likely to also have occurred in these areas. In this contribution, we analyze to what extent the emissions reductions are detectable in routine measurements of air quality in Central Europe. Our area of interest is the Fichtelgebirge, NE Bavaria, a mountainous range in the center of Europe, reaching peak altitudes of just over 1000 m a.s.l. This area is strongly affected by forest decline phenomena (SCHULZE et al., 1989). It is of great interest in our ecosystem research studies to know whether a significant change in air quality has occurred over the years. 2. DATA BASE Most of the routine measurements of air quality in Germany are conducted in the centers of cities. These data are used to decide if the levels of air pollution are likely to affect the health of the human population. We are, however, more interested to know whether the vegetation in rural areas is likely to be affected by air pollution or not. As part of a program to study the forest decline phenomena in Northern Bavaria, one station has been set up in 1985 at a remote, forested site close to the village of Warmensteinach in the "Fichtelgebirge" at an altitude of 575 m a.s.l. After operating at that site until 1993, the station has been moved northward to the research site "Waldstein", at 765 m a.s.l. at a distance of 18 km from the earlier site in Warmensteinach. The move led to a gap in the data series of 8 months duration. In combination, a 13 year time series of air quality measurements is available for the Fichtelgebirge region. However, there are some limitations concerning the interpretability of the data set. First, it is not clear whether the move led to any systematic bias between the two data subsets and secondly, the locations
of both stations are in small forest clearings, so that the respective micrometeorological conditions are heavily modulated by the local topography and vegetation. Within our data analysis, we want to estimate where air masses with high or low trace gas concentrations originate. For example, we want to quantify if, in a given time period, air masses of very high SO2 concentrations come from Eastern Germany, located to the North of our research station, from the Czech Republic in the East, or from western Europe, located in the West and South. However, it is not feasible to calculate a large number of backward trajectories on a daily or even more frequent basis. We made use of the meteorology data that were acquired by the Bavarian State Ministry for State Development and Environmental Affairs (StMLU) and by the German Weather Service at their stations outside the cities near the Fichtelgebirge mountains (Bayreuth, Kulmbach, Naila, Hof, Selb, Arzberg). We define a mean wind vector of the Fichtelgebirge, estimating the regional flow field to the best possible extent, by averaging all available wind vectors from the city stations for every hour of measurement. Although the physical meaning of a mean vector is limited because different stations exhibit different wind directions at a given time, we suggest that this vector is a much better representation of the regional flow pattern than the winds measured at the forest stations. Hereafter, the hourly averaged spatial mean of the wind vector will be used to interpret any measured trace gas concentrations with respect to the associated wind directions. Further, we created a synthetic data set for SO 2 that represents the SO2 concentrations in the cities at the upwind side of the Fichtelgebirge mountains. For situations with winds from NW to NE (mean wind direction between 315 and 45 degrees), the mean SO 2 concentrations of the cities of Naila and Hof were used. For easterly winds (45 through 135 degrees), the averages of the stations Selb and Arzberg were used, and for the remaining data (wind from 135 through 315 degrees), we used the data from Bayreuth and Kulmbach. In our region, high SO2 concentrations occur in episodes that last between a few hours and several days. It appears from a visual inspection of the original data sets that episodes of high SO2 occur simultaneously at the forest sites and in the cities upwind of the Fichtelgebirge, respectively. Further, it appears that the two forest sites reflect the pattern of the SO2 concentrations in the Fichtelgebirge in a similar manner. They have indistinguishable Pearson correlation coefficients with the synthetic
data set for the respective periods (r2=0.50 and r2=0.46 for Warmensteinach and Waldstein, respectively). The SO2 data from the forest sites (see Fig. 1, solid line, for a frequency distribution) cover a wide range of up to 250 ppb. More than half of the data are at the lower edge of the scale between 0 and 4 ppb. This leads to an extremely skewed distribution (nondimensional skewness = 4.907, which is about by a factor of 400 larger than the skewness of a Gaussian distribution of the same sample size) which can not be appropriately described with a parametric distribution. Further complications arise from the fact that the low mixing ratios are in many cases close to or below the detection limit of the employed analytical instruments. For the analysis of NOx and O3, we rely exclusively on the data from the forest sites in Warmensteinach and at the Waldstein because the city data of NOx and O3 are governed by the local road traffic, which injects NO into the atmosphere that reacts with O 3 within a few minutes. NOx measurements were conducted at Warmensteinach only between September 1988 and December 1990. At the Waldstein site, NOx data are available for the entire period between 1994 and 1997. The O3 data of the forest sites cover the entire period 1985 - 1997 with some large (1 to 2 months duration each) gaps in the data records between 1985 and 1993. At the Waldstein station (1994 through 1997), the O 3 data record is almost complete. All trace gas data are given in units ppb, which means a volume (or molar) mixing ratio of 10-9. Note that other publications and data bases sometimes use concentrations units such as µg ⋅ m-3. In many cases, the concentrations refer to a volume of air under standard conditions (1013 hPa, 298 K). For these conditions, a mixing ratio of 1 ppb equals a concentration of 2.6 µg ⋅ m-3 in the case of SO2 and 2 µg ⋅ m-3 in the case of O3. 3. TIME SERIES INVESTIGATIONS A variety of investigation tools yield information about periodicities, persistence, instationarities, and trends in the data sets. The characterization of the data sets for SO2, NOx, and O3, is guided by two questions: First, are there indications for a significant bias induced by the transfer of the forest measurement station from Warmensteinach to the Waldstein site in 1993/1994? And secondly, are there any observable long-term
trends? For an evaluation of the first question, we calculate the autocorrelation functions (section 3.1), Hurst statistics (3.2), two complexity measures (3.3), and recurrence quantification parameters (3.4) for subsets from Warmensteinach and Waldstein, respectively. For a clarification of the second question, we also use the results from sections 3.1 through 3.4 for a characterization of the time series with respect to the presence of periodicities and long-term as well as short time behavior, and, in addition, we perform a trend analysis in section 3.5. Because of the gap within the data sets that resulted from the move of the station in 1993/1994, it is not meaningful to calculate any statistical measures that require equidistant data for the entire period of investigation. We analyze two representative data subsets of approximately equal length from Warmensteinach (1985 - 1987) and from the Waldstein (1994 - 1997), respectively. These measurement periods are also not completely gap-free. Nevertheless, we tried several different interpolation schemes to fill the gaps in the data sets and found that our analyses were not significantly affected. Therefore, we ignored the data gaps in the following analyses.
3.1 AUROCORRELATION FUNCTIONS One of the standard tools for linear time series analysis is the autocorrelation function (ACF) r (k ) of a data set x (ti ) obtained from the autocovariance
φ (k ) =
1 N −k ∑ ( x(ti +k ) − x )( x(ti ) − x ) N − 1 i =1
and r ( k ) = φ (k ) / φ (0) = φ ( k ) / σ x2
where k is the time lag considered, N the length of the series, x its overall arithmetic mean, and σ x2 its variance.
The ACF quantifies linear dependencies between values within a data set; all r (k ) lie between –1 and +1, and values significantly above zero denote correlation at the given lag. A negative r (k ) indicates significant anticorrelation. Existing periodicities are exhibited by regular peaks in r (k ) ; their heights may be used to estimate their importance relative to other (e.g. non-periodic) correlations. A finite memory of the series leads to a gradual decline of the envelope of the ACF. The autocorrelation functions for our series are shown in Figure 2. There is a well pronounced seasonality for O3 at the Waldstein site. This phenomenon is mainly due to the fact that O3 is predominantly produced during summer due to high NO 2 photolysis frequencies. However, the amplitude of the autocorrelation function becomes less pronounced for longer time lags (>1 a), and after 3 years, the autocorrelation coefficients reach only values of around 0.1. This is an indication that the memory of the O 3 signal is finite, i.e, an autocorrelation length is calculable, and that the periodicity is not extremely regular. It may also point to a trend in the data set. On the other hand, the plausible finite size effects at large time lags usually lead to large fluctuations in the ACF that are not observed here. For the Warmensteinach site, the seasonality is also indicated; however, in this case, the measurement period is too short to lead to definite conclusions. The seasonality for NOx is also clearly established. As the emissions of NOx are typically not higher in the winter months than during the summer, this result is, at first glance, surprising. We assume that the origin of the seasonal pattern of NOx lies in the different meteorological conditions during winter rather than in different NO x emission conditions: In winter, the near-surface atmospheric boundary layer is generally less turbulent and less deep. Therefore, the NO x emssions from surface sources are less well mixed within the lowest layer of the atmosphere and therefore establish higher concentrations of NOx at a site on the ground. The autocorrelation of SO2 also reaches significant values on an annual cycle, as would be expected from the observation that episodes of high SO2 are more frequent and more pronounced during winter than during summer. However, the autocorrelation of SO 2 phases out rather quickly (within about two years), pointing to a much more pronounced finite memory range than in the O3 case, and also to the presence of a trend.
3.2 HURST ANALYSIS Further insight into the extension and importance of memory effects and the long-term structure of our atmospheric time series is gained by calculation of the Hurst exponent H. This measure quantifies the extension of periods with systematic deviations from the overall mean, which is also called the persistence of the time series. If the value range of a time series scales with the length of the time series, the Hurst effect is present. The best fit value for H found typically in experimental data sets is 0.5 ≤ H ≤ 1.0; the lower limit corresponds to ordinary Brownian motion (low persistence), the upper one to very high persistence; values in between correspond to the so-called fractional Gaussian noise (RODRÍGUEZ-ITURBE and RINALDO, 1997). There are different techniques to estimate H. Our calculation was performed with the rescaled range or R/S statistics (MONTANARI et al. 1997). In all our analyzed data sets, persistence is clearly present (H > 0.85). The H values for O3 from Warmensteinach (0.92 ± 0.03) is remarkably similar to the one from the Waldstein (0.93 ± 0.01). Also for SO2, the H values from the two sites are very similar (0.87 ± 0.03 for Warmensteinach and 0.86 ± 0.02 for Waldstein, respectively). On the other hand, for a single research site, the difference between the two gases is highly significant (e.g., 0.93 for O3 vs. 0.86 for SO2 at the Waldstein site). We conclude that the Hurst exponent is a good characterization of the respective gases and that the characteristics are almost identical for a single gas (e.g., O3) in the two data subsets. There is no indication of a systematic bias in the two data sets. The particularly high H exponents of O3 probably originate in the fact that high O3 typically occurs during photochemical episodes in summer that typically last several days. For SO2 and NOx, the Hurst exponents are also quite high (0.88 ± 0.03 for NOx at the Waldstein site), again emphasizing the episodic nature of events with very high concentrations. 3.3 COMPLEXITY MEASURES Short term structures of the time series are investigated with the aid of complexity measures. This serves to elucidate the relationship between randomness of a data set (in many cases correlated with noise level) and its "true" complexity (BAR-YAM 1997). The use of this concept refers to the difficulty to describe the temporal structure in a compact
way. When quantified, complexity should be low for nearly constant or slowly-varying signals, and should also be low for nearly completely random sequences as the statistical description of a white–noise process (a completely unstructured process with no correlations among consecutive values, e.g., an evenly distributed random distribution) is easy. Calculation of these measures provides information about the optimal temporal resolution for measurements of a variable (LANGE et al. 1998, LANGE 1999). The interplay between driven and independent parts, or between underlying processes of different time scales, determines the quantification of randomness and complexity. Technically, complexity measures are calculated on the basis of a discretized version of the original series, the so called symbol sequences (WACKERBAUER et al. 1994). Small groups of consecutive values within these sequences are called words, the length L of which has to be fixed (we have chosen L = 4 for all investigations here). The method quantify the short-term correlations within the range prescribed by the word length L in an average manner. Complexity is quantified as fluctuation complexity (BATES AND SHEPHARD, 1993), which vanishes both for constant as well as random data and has a maximum in–between constant and random data. Our preferred randomness measure is the Mean Information Gain (MIG, W ACKERBAUER et al. 1994). The MIG is minimal for constant and maximal for random sequences. We present the calculations in complexity diagrams where MIG is the abscissa and the fluctuation complexity the ordinate. The curves in Figure 3 were obtained by varying the temporal resolution by aggregation (1 – 48 h). As expected, the randomness measure increases as more and more short–term correlations are averaged out, whereas the fluctuation complexity first increases and then decreases. The difference between SO2 and O3 at high aggregation level is obvious, as is the remarkable similarity between the two subsets. This indicates that the small–scale structure of the data sets is not influenced by the move of the research site from Warmensteinach to Waldstein. The distance of the randomness/complexity values from the theoretical limit (obtained for a Bernoulli sequence) is larger than for typical hydrological variables (L ANGE 1999). At the same level of randomness, less complexity is found in our data sets for atmospheric trace gases than for water signals (i.e., runoff from catchments). The fact that the O3
signal becomes insensitive with respect to randomness for sufficiently high aggregation level stems from the daily cycle of photolysis; for a strictly periodic signal with period P, the MIG is constant once the words span more than the period ( L ≥ P ). In our case, with L = 4 and hourly resolution, this is the case for aggregation level 24 / 4 = 6, confirmed for both measurement stations. For SO 2, no daily cycle is visible. Note that the annual cycle is far from being detectable by this method as the maximal word length is limited by the need to obtain saturated word frequency statistics. Maximal complexity values are obtained for both gases at an aggregation width of 2–3 hours. 3.4 RECURRENCE QUANTIFICATION We used the technique of recurrence plots (ECKMANN et al. 1987) and their quantification (ZBILUT et al. 1998) to estimate the degree of determinism (DOD) in the respective time series. Recurrence plots are a visualization of recurrent patterns at arbitrary locations within a time series. They are constructed from a time series of length N by firstly building embedding vectors (d consecutive values put together), and secondly calculating vector distances between them for each pair. In this way, a matrix is built and matrix elements below a predefined threshold distance r are considered as recurrent points. Deterministic (sub-)structures in the recurre nce matrix are revealed by short line segments parallel to the main diagonal. Defining a minimal line length to avoid counting accidental lines, one calculates the ratio of recurrent points which are members of lines and the total number of recurrent points to obtain the DOD. As the DOD value clearly depends on the threshold value and minimal line length chosen, one should not interpret the absolute values quantitatively. However, if we choose the same parameter set for O 3 and SO2 and consider the DOD values of consecutive (overlapping) windows of fixed length, relative degrees lead to meaningful conclusions. Figure 4 clearly demonstrates that the DOD characterizes the variable rather than the measurement period. For SO 2, the DOD is generally high (low noise level). For the Warmensteinach period, it seems to be stationary and covaries with the percentage of recurrent points, indicating the absence of structural instationarities. For the Waldstein data set, there are two remarkable exceptions to this covariation of unknown origin, and a decreasing DOD trend towards the end of the period. This may be due to the fact that
the declining SO2 concentrations also reduce the distance to the noise floor (measurement resolution). The O3 data are similar for the two stations, but drastically different from the SO 2 signal: the O3 DODs are quite low, the signal is thus substantially noise-corrupted, which is plausible from visual inspection of the time series, and a seasonal cycle is present, with higher values for DOD when O3 concentrations are low. 3.5 TREND ANALYSIS Trend statistics was calculated using percentiles of the measured mixing ratios for every year between 1985 and 1997 and applying the non-parametric Mann-Kendall test for the presence of trends (HOLLANDER and WOLFE, 1973; GILBERT, 1987) and Sen’s method (GILBERT, 1987) for determining the slopes of the trends. The results are summarized in Figure 5. We found decreasing trends of the SO 2 mixing ratios at the forest sites. All trends are highly significant (α ≤ 0.05) and we note that the highest percentiles decreased more steeply and with higher significance than the lower percentiles. For the forest sites, the decrease of the SO2 mixing ratios is further visualized in Figure 1: For high mixing ratios (> 13 ppb, or > 1.1 on the logarithmic abscissa), the relative frequency line of the Warmensteinach data set is well above the corresponding line of the entire data set and even further above the Waldstein line, indicating that higher SO 2 mixing ratios were much more frequent in the earlier data. In the cities, the SO2 were higher (about by a factor of 2, Fig. 5), but also decrease more rapidly within the 13 year period. The cities/forest ratios decreases over the years: after exhibiting ratios between 2 and 3 in the mid–1980's, the ratios seem to level out close to unity in the mid–1990's (0.996 ≤ ratio ≤ 1.052 in the year 1997). Figure 6 shows the SO2 wind roses for the earliest years of our measurements and for the latest years at the forest sites. In Warmensteinach (1985 - 1987), the highest concentrations are typically found when the winds were from the North and Northeast. 95% percentiles around 100 ppb were found for wind direction averages of 0°, 30°, and 60°, respectively. For the later data from the Waldstein site, the most important feature seems to be the strong decrease of the SO2 mixing ratios as compared to the earlier data, as it would be expected from the results as presented in Figure 5. It becomes
additionally clear from Figure 6 that the SO 2 mixing ratios decreased for all wind directions. The wind direction with the highest SO2 mixing ratios has turned clockwise in the later data sets compared with the earlier ones: at the Waldstein site, the SO2 maxima were clearly associated with winds from 60°, while in Warmensteinach the maximum covered a broader angle between 0° and 60°, with the maximum at 30°. This result indicates that the decrease of the SO2 was stronger for northerly winds than for northeasterly and easterly winds. Statistical analysis confirms this observation of a much more significant (α ≤ 0.002) and a much steeper (-3.26, -2.73, -2.33 ppb a-1 for the 95%, 90%, and 80% percentiles, respectively) decrease for SO 2 associated with northerly winds than for easterly winds (0.082 ≤ α ≤ 0.092; slope -1.54, -1.06, and -0.68 ppb a-1, respectively). Here we observe a direct effect of the decrease of the SO 2 emissions between the mid 1980's and the mid 1990's which was much more efficient in eastern Germany, located to the North of our site, than in the Czech Republic to the East (Umweltbundesamt, 1997; EMEP, 1997). The O3 mixing ratios exhibit a highly significant increasing trend at our forest sites (Fig. 5). The high percentiles of the O3 distribution increase more intensively than the low percentiles (e.g., medians). This indicates that it is the frequency and/or intensity of episodes of high O3 that dominates the general increase of O3 at our sites. For the nitrogen oxides, our data base is much weaker, and no significant trends could be found. 4. CONCLUSIONS We analyzed a 13-year time series of hourly SO 2, NOx, and O3 data from forest sites in the Fichtelgebirge mountains, NE Bavaria, Germany. We applied extensive statistical tests (sections 3.1 - 3.4) to find indications of whether the move of the forest measurement site to a different location in 1993/1994 introduced any systematic bias into our data sets. All tests, operating independently from each other and referring to different characteristic qualitites of the data sets, indicated that there are no systematic differences between the data sets from the two sites. We conclude that it is appropriate to construct a combined 13-year time series. We obtain further support for the applicability of this approach by comparing the combined data set from the forest sites with a synthetic SO2 data set representing the urban stations at the respective upwind side of the Fichtelgebirge mountains.
Most of the time, the cities show higher SO2 mixing ratios than the forest sites. City stations are, in terms of atmospheric transport time, always closer to SO 2 emission sources (large industry and power plants but also domestic lignite burning and small industry) than the forest sites. Any SO2 being advected from the upwind side of the Fichtelgebirge mountains to our forest sites is subject to dilution, deposition to the surface, and to gas and liquid phase oxidation. These processes are very likely to result in a decrease of the SO2 concentration during their transport from the cities to the forest sites. However, the differences in SO2 concentrations in cities as compared to the forest site decreased strongly between 1985 and 1997. This decrease probably reflects the fact that the reduction of sulfur emissions in cities was more effective than for the average of all emissions. The SO2 shows a particularly strong decrease in both the forest and city data sets between 1987 and 1988 due to the enforcement of air pollution control strategies in West Germany, while the breakdown of the East German enonomy after opening of the German border and resulting emission reductions between 1990 and 1991 are not reflected in our data sets. Overall, the emissions of SO 2 were reduced by about or 61% between 1985 and 1994 (Umweltbundesamt, 1997). The agreement of this decrease with the decrease of the 95%, 90%, and 80% percentiles of SO2 in the forest data set (61%, 64%, and 36%, respectively) is convincing. We conclude that the reduction of the SO2 emissions lead to a one-to-one response of the high (95% and 90% percentiles) SO2 mixing ratios at our forest sites. Nitrogen is emitted into the atmosphere mainly as NO and NH3. Atmospheric transport, reactivities, and deposition mechanisms of these species and their products (NO 2, HNO3, particulate NO3-, particulate NH4+, and others) are very different from each other. Our data base is too weak to draw quantitative conclusions, but it should be stated that the emission reductions of 29% in Germany between 1985 and 1994 are not reflected in the mixing ratios at our forest site. Also, no preference of higher mixing ratios has been found according to certain wind directions (such as for SO 2, Fig. 6). The results are in general agreement with studies from other sites in Europe, where reductions of atmospheric deposition of sulfur compounds was found but not for nitrogen compounds (AVILA, 1996).
Our O3 data show a distinct increase over the 13 year period of observation. The 95% percentile increased by almost 2 ppb ⋅ a-1. The atmospheric processes governing the production and transport of O 3 are highly complex and nonlinear and it is therefore not clear from our analysis what are the underlying causes for this increase in O 3. Any threshold or limit values of the German and European Union authorities for SO 2 and NOx (e.g., EU threshold of 100 µg ⋅ m-3 SO2 as 24-hrs-mean) were not even closely approached at any time at our forest site. For O 3, however, the EU threshold for the protection of vegetation (200 ppb one-hour-mean) was frequently exceeded. Ozone also exceeds the critical O3 level for forest protection (exposure over the threshold value of 40 ppb during the summer months and during the daylight hours, AOT40, Fuhrer et al., 1997) of 10 ppm ⋅ h since 1991, with values between 17.3 and 21.5.
Acknowledgements: This work was supported by the German Bundesministerium für Bildung, Wissenschaft, Forschung und Technologie (BMBF) through grants Nos. PT BEO 51-0339476B+C. We are indebted to the Bavarian Landesamt für Umweltschutz and the Deutscher Wetterdienst for supplying us with data from their monitoring networks. Thanks to T. Ferdelman for language-editing of the manuscript.
REFERENCES AVILA, A. (1996): Time trends in the precipitation chemistry at a mountain site in northeastern Spain for the period 1983–1994. Atmos. Environ. 30, 1363-1373. BATES, J.E. and SHEPHARD, H.K. (1993): Measuring complexity sing information fluctuations. Physics Letters A 172, 416-425. BAR-YAM, Y. (1997): Dynamics of Complex Systems. Addison-Wesley, Reading. ECKMANN, J.P., KAMPHORST, S.O. and RUELLE, D. (1987): Recurrence plots of dynamical systems. Europhysics Letters 4, 973-979. EMEP MSC-W Report 1/97 (1997): Emissions, dispersion and trends of acidifying and eutrophying agents. Norw. Met. Inst., PO-Box 43-Blindern, 0313 Oslo 3, Norway. GILBERT, R. 1987. Statistical Methods for Environmental Pollution Monitoring. Van Nostrand Reinhold. Melbourne. FUHRER, J., SKÄRBY, L. and ASHMORE, M.R. (1997) Critical Levels for ozone effects on vegetation in Europe. Environ. Pollut. 97, 91-106. HOLLANDER, M., and WOLFE, D.A. 1973. Nonparametric Statistical Methods. John Wiley & Sons. New York. LANGE, H., NEWIG, J. and WOLF, F. (1998): Comparison of complexity measures for time series from ecosystem research. Bayreuther Forum Ökologie 52, 99-116. LANGE, H. (1999): Are Ecosystems Dynamical Systems? Journal of Computing Anticipatory Systems (in press). MONTANARI, A., ROSSO, R., and TAQQU, M.S. (1997): Fractionally differenced ARIMA models applied to hydrologic time series: Identification, estimation, and simulation. Water Resources Research 33, 1035-1044. RODRÍGUEZ-ITURBE, I. and RINALDO, A. (1997): Fractal River Basins. Cambridge University Press. SCHULZE, E.-D., LANGE, O.L., and OREN, R. (Eds.) 1989. Forest Decline and Air Pollution. Ecological Studies 77, 475 pp. UMWELTBUNDESAMT, (Ed.), 1997. Daten zur Umwelt. Erich Schmidt Verlag. Berlin. WACKERBAUER, R., WITT, A., ATMANSPACHER, H., KURTHS, J., and SCHEINGRABER, H. (1994): A Comparative Classification of Complexity Measures. Chaos, Solitons & Fractals 4, 133 - 173. ZBILUT, J.P., GIULIANI, A., and WEBBER Jr., C.L. (1998): Recurrence quantification analysis and principal components in the detection of short complex signals. Physics Letters A 237, 131-135.
60 Warmensteinach 1985 - 1987 all forest data 1985 - 1997 Waldstein 1994 - 1997
d (relative frequency / %) / d (log SO2 mixing ratio / ppb)
50
40
30
20
10
0 0.6
1.0
1.5
2.0
log (SO2 mixing ratio / ppb)
Fig. 1: Frequency distributions of the hourly SO 2 data (all data and partial data sets) from the forest sites. The lower end of the distributions (SO 2 mixing ratio < 4 ppb, or log(SO2) < 0.6) are not shown because they are influenced by the detection limits of the employed analytical instrumentation.
time lag / a 0
1
2
3
0.6
Waldstein 1994 - 1997 O3
0.5 0.4 0.3
SO2
NOx
0.2 0.1
autocorrelation coefficient r
0.0 -0.1 -0.2 -0.3
0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 -0.4
Warmensteinach 1985 - 1987
O3
0
SO2
1
2
time lag / a
Fig. 2: Autocorrelation functions (ACF) of trace gas data at the two forest sites in the Fichtelgebirge mountains. Any data points above the dashed line indicate significant positive autocorrelation for the respective time lag. NO x was not measured in Warmensteinach between 1985 and 1987.
Mean Information Gain 0.2 2.0
0.3
0.4
0.5
0.6
0.7
SO2
Warmensteinach Fluctuation Complexity
1.8
O3
1.6 1.4 1.2 1.0
1.8 Fluctuation Complexity
0.8
Waldstein
SO2 O3
1.6 1.4 1.2 1.0 0.8 0.2
0.3
0.4
0.5
0.6
0.7
0.8
Mean Information Gain
Fig. 3: Fluctuation complexity (FC) versus mean information gain (MIG) of the SO2 and O3 from Warmensteinach and the Waldstein site. The solid lines describe the theoretical upper limit of the FC.
100 80
1985
1986
1995
1987
SO2
SO2
O3
O3
1996
1997
1996
1997
60 40 20 100 80 60 40 20 0
1985
1986
1987
Warmensteinach
% Recurrence,
1995
Waldstein
% Determinism
Fig. 4: Percent recurrence and percent determinism of the SO 2 and O3 data from Warmensteinach and the Waldstein site, respectively. The x-axes are the end times of the time windows (window lengths: 83 days).
100
1985
1987
80
1989
1991
1993
1995
1997
SO2 forest sites / ppb
-1
-1.84 ppb a ; α = 0.015 -1 -1.10 ppb a ; α = 0.021 -1 -0.50 ppb a ; α = 0.025
60 40 20 0 100
-1
-4.71 ppb a ; α < 0.001 SO2 cities / ppb -1 -2.98 ppb a ; α < 0.001 -1 -2.07 ppb a ; α < 0.001
80 60 40 20 0 100 80
O3 forest sites / ppb +1.90 ppb a-1-1; α = 0.003
+1.70 ppb a ; α = 0.015
60 40 20
-1
0 1985
1987
1989
+1.40 ppb a ; α = 0.021 -1 +0.70 ppb a ; α = 0.034 1991 1993 1995 1997 year
95% percentile 80% percentile
90% percentile 50% percentile (median)
Fig. 5: Annual percentiles of the SO2 and O3 mixing ratios between 1985 and 1997. Significant trends and the respective significance levels (α) of the various time series are given.
Fig. 6: Wind roses of the SO2 mixing ratios (ppb) for the forest sites. Bold diagonals indicate a sectorization of the wind rose according to estimated regional origin of the respective air masses.
