Spatial gradients in ratios of atmospheric trace gases: a study stimulated by experiments on bird navigation

Hans G. Wallraff; Meinrat O. Andreae

doi:10.3402/tellusb.v52i4.17084

Spatial gradients in ratios of atmospheric trace gases: a study stimulated by experiments on bird navigation

Wallraff, Hans G.; Andreae, Meinrat O. 2000-01-01 00:00:00 T ellus (2000), 52B, 1138–1157 Copyright © Munksgaard, 2000 Printed in UK. All rights reserved TELLUS ISSN 0280–6509 Spatial gradients in ratios of atmospheric trace gases: a study stimulated by experiments on bird navigation By HANS G. WALLRAFF1,* and MEINRAT O. ANDREAE2, 1Max Planck Institute for Behavioural Physiology, D-82319 Seewiesen, Germany; 2Max Planck Institute for Chemistry, Biogeochemistry Department, PO Box 3060, D-55020 Mainz, Germany (Manuscript received 9 June 1999; in ﬁnal form 22 December 1999) ABSTRACT Numerous experiments with homing pigeons and other birds strongly suggest that birds dis- placed to unfamiliar remote areas are able to determine their position relative to home by deducing relevant information from atmospheric trace gases perceived by olfaction. These ﬁnd- ings induced the hypothesis that ratios between several airborne compounds show roughly monotonic spatial gradients, diVerently in diVerent directions, over distances of some hundreds of kilometres. To test this hypothesis, 192 air samples were collected, successively in 3 summers, at 96 sites regularly distributed over an area covering a radius of 200 km around Wu ¨ rzburg, Germany. Statistical analysis of the gas chromatographic measurements on these samples revealed that such gradients in the ratios between a number of omnipresent hydrocarbons do in fact exist. The gradients are noisy, but not beyond the range that is compatible with the homing behaviour of pigeons which is noisy as well. The directions of the gradients are remark- ably robust against changes of weather, especially of winds. Winds, however, shift the levels of ratios in the whole area without dramatically changing the directional relationships. A system- atic angular correlation between variations in space and variations caused by winds could theoretically be utilized by birds for navigational purposes. Our analysis dealt mainly with the most abundant anthropogenic hydrocarbons, which are the best-suited tracers to detect spatio- temporal distribution patterns. It is very likely that equivalent patterns exist in naturally emitted volatile compounds as well, given that they are subject to similar variability in the distribution of sources and sinks and similar transport patterns. 1. Introduction a medium they had never expected to be of interest in this context: the atmosphere. Over more than 1.1. T he problem 25 years a large body of evidence has accumulated that strongly suggests that homing pigeons and This investigation was stimulated by navigating other birds are able to deduce spatial information birds. After many years of unsuccessful searches from atmospheric trace gases perceived by olfac- for the physical basis of their ability to home from tion (Papi, 1989, 1991; WallraV, 1990a,b, 1996; unfamiliar distant areas and after exclusion of WallraV et al., 1995; Able, 1996; Bingman, 1998). deductively reasonable sources of positional It is still enigmatic, however, how this might be information, such as the geomagnetic ﬁeld or possible. Doubts about the feasibility of olfactory astronomical clues, researchers were forced by navigation (Schmidt-Koenig and Ganzhorn, 1991; experimental ﬁndings to focus their attention on Wiltschko, 1996; Gould, 1998) ﬁnd immediate belief, because intuitively the atmosphere appears, * Corresponding author. e-mail: [email protected] due to its temporal instability, utterly unsuited as Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1139 a source of spatial information. This preconception which is not individually peculiar but diVers has seemingly been supported by negative reports gradually from the compositions in the neigh- on the suitability of atmospheric trace substances bouring areas. for navigational purposes (Becker and van Raden, (2) Homing of pigeons on an olfactory basis is 1986; Waldvogel, 1987, 1989; Ganzhorn and possible even if the birds lack continuous contact PaVrath, 1995). However, by only considering with environmental air during the outward trip to absolute concentrations of single compounds sep- the release site. Thus, it should be possible to arately, these studies started from restricted pre- determine positions of distant sites relative to the mises which, a priori, could hardly be expected to home site by analyzing ambient air only at these raise positive results. Given the extent of the sites themselves. results on homing, which, as a whole, cannot (3) Pigeons home successfully from sites that plausibly be explained in any alternative way, we may be a hundred or more kilometres away from should trust the pigeons rather than our limited sites they have previously visited. Thus, the knowledge and imaginative competence. Solutions pigeons need not develop a topographical map of can only be expected from a more thorough odours by exploring a large area. Instead, atmo- empirical search for persistent patterns of atmo- spheric conditions appear to allow spatial extra- spheric tracers, particularly with respect to their polations from one site ( home) to an indeﬁnite suitability as a source of positional information. number of other sites in a large surrounding area. In this study, we present the result of the ﬁrst Extrapolations are possible only if there is some investigation of this kind. We show that the atmo- predictable quantitative connection between con- sphere does contain surprisingly persistent spatial ditions at diVerent sites, i.e., if the spatial distribu- structures in the relative proportions of chemical tions of particular airborne trace compounds tracers, structures which potentially provide a include some kinds of fairly monotonic gradients suitable basis for avian navigation. Studies on over longer distances. pigeon homing have thus inspired a search for (4) Since we know that pigeons can ‘smell’ their mesoscale chemical gradients that had not been position before they start ﬂying, they must already previously noticed in the atmosphere. have some knowledge about the directions of gradients without scanning a larger area upon 1.2. Scope of search deduced f rom homing release. Experimental studies suggest that they experiments with pigeons gain this knowledge during their long-term stay at the home site by correlating olfactory conditions Previous experiments with homing pigeons led with the direction of the wind. Normal homing to a working hypothesis predicting that certain behaviour can be observed only in pigeons that properties of the atmospheric chemical environ- were exposed to natural winds at the home site. ment make it suitable as a source of spatial The birds are disoriented and fail to return when information. These hypothetical properties were they previously have been kept in an aviary deduced from the following experimental ﬁndings, screened from winds. Their departure directions conclusions and associated considerations, which can be predictably manipulated by manipulating are all based on the general conclusion that pigeon wind directions in the home aviary. The necessary homing from unfamiliar areas relies on olfactory directional reference is available in the form of a signals (for review of the results themselves, critical sun compass which is known as a part of the valuations, debated interpretations and detailed pigeons’ homing mechanism. references, see Papi (1989, 1991) and WallraV (5) For the pigeons, long-term signal averaging (1990a,b, 1996 )). is possible at one position, the home site. When (1) Pigeons home equally well with both head exposed to the local air at a distant site, they must and tail winds. Thus, they cannot navigate using be able to deduce their position in relation to a chemical signal that is blown directly from the home site to their current position. If, nevertheless, home by comparing actual olfactory conditions pigeon homing is based on odours, the home site with remembered home conditions. The time avail- should be characterized by a certain qualitative able for this comparison is usually limited to, at and quantitative composition of compounds most, a few hours. A period of only 5–10 min, Tellus 52B (2000), 4 1140 h. g. wallraff and m. o. andreae however, seems to be insuYcient to reach full didates also from a physiological point of view, performance. because organisms usually have diYculties in (6) Spatial information deduced from atmo- quantifying absolute intensities. However, animals spheric trace compounds need not be precise. and humans can detect small variations in the Pigeons do usually not ﬂy the shortest way straight proportional composition of a set of odorous home, many fail to home at all, and the routes of substances even with varying levels of absolute a sample of individually ﬂying birds show consid- concentrations. At least in humans, diVerent erable stochastic noise. Orientation performances quantitative proportions between compounds pro- as shown by pigeons can be achieved, on a proba- duce subjectively diVerent qualitative odours. bilistic basis, by evaluating noticeably unreliable The above results and considerations led to the environmental signals. Thus, we need not expect following working hypothesis (WallraV, 1989b, to ﬁnd perfectly monotonic gradients in the atmo- 1991, 1996). A number of trace compounds (min- sphere; considerable noise can be tolerated. imum three), permanently present in the ground- (7) Spatial information deduced from atmo- level air, should systematically vary in their con- spheric trace compounds need not be stable in centrations relative to each other over distances time, as also pigeon homing is temporally highly of several hundreds of kilometres (at least in variable in scales of hours, days, years, and accord- Germany), diVerently in diVerent directions. As a ing to seasons. Performances in winter are much result, any location inside a wide area should be lower than in summer (as are emissions of biogenic characterized by a unique proportional composi- trace gases). tion of concentrations of these compounds (see (8) The spatial range within which pigeons can background stippling in Fig. 1 in WallraV, 1996). deduce their positions relative to home from local This prediction need not be perfectly met; consid- air is fairly large but not unlimited. It seems to erable noise is allowed. To a smaller or larger vary from less than 100 to at least 700 km, appar- degree, the pattern of gradients should systematic- ently depending on geographical ( possibly oro- ally vary with varying directions of the wind. graphical ) conditions. Spatial variations and wind-dependent variations (9) Homing ﬂights of pigeons usually take place should be integratable into a common system. at altitudes of less than 300 m, mostly less than Currently in Central Europe, where many of 100 or 50 m above ground (Hitchcock, 1952; the pigeon homing experiments were conducted, Wagner, 1970). Moreover, pigeons can pick up the most prominent trace gases accessible to gas navigationally useful olfactory information at chromatographic analysis are volatile organic heights of less than 5 m. Air above open land has compounds ( VOCs), mainly hydrocarbons, been demonstrated to contain more reliable released into the atmosphere by human activities, information than air inside forests or between such as vehicular emissions, gasoline production other vegetation. Thus, trace substances to be and handling, and solvent evaporation (Piccot analyzed can be collected close to ground level, et al., 1992; Singh and Zimmermann, 1992; Atlas but air from above open landscapes should be et al., 1993; Fujita et al., 1995; Clarke and Ko, preferred. 1996; Mukund et al., 1996). Thus, the atmosphere (10) Pigeons deduce navigational signals from investigated today diVers considerably from the airborne molecules in the gas phase. Removal of atmosphere in which avian navigation systems aerosol particles by ﬁltration does not interfere evolved. However, this apparent disadvantage may with navigation, whereas scrubbing by charcoal provide a methodological advantage. Due to their eliminates positional information. higher concentrations, many anthropogenic VOCs can be more reliably detected and quantiﬁed than natural compounds. Our hypothesis includes that 1.3. Analytical approach spatiotemporal distributions of natural and A priori, we did not search for spatial gradients anthropogenic compounds follow similar rules in the absolute concentrations of single com- because they are determined by the same atmo- pounds. Their temporal variability is enormous. spheric processes. If this is true, it must be easier Much less variable are proportional relations to decipher these rules in the latter class of com- between compounds. They are the favoured can- pounds. The following analysis deals with the Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1141 proportional relationships among all substances 12 peripheral sites in succession, 3 on each arm of detected by the applied method without regarding a rectangular cross, with 60 km between the sites their chemical nature and origin. on each radius. Four such crosses at angles of about 22.5° to each other, each including 2 sets of alternating sites on each cross, resulted in 96 sites 2. Material and methods (Fig. 1). In 2 successive rounds, each consisting of 8 trips, each site was visited twice in diVerent 2.1. Collection of material years. In addition, two samples were taken at the centre of the ﬁeld, at intervals of 2–3 days, during Samples of atmospheric trace gases were col- each trip (=32 samples). Sampling sites were in lected using small adsorbent tubes (Supelco open or fairly open rural (rarely semi-rural ) coun- Carbotrap 400, containing Carbotrap F, C, B and try, preferentially on hills rather than in valleys Carboxen 569). The sampling device was set up and usually at some distance from larger forests. on the roof of a mini-van with the inlets of the The centre of the sampling grid was identical with sampling tubes about 4.25 m above the ground. the position of a pigeon loft from which many Sampling duration was 2 h and sampling volume homing experiments had been conducted inside 48 l of air. Before and after sampling, the tubes the same area. were stored in liquid nitrogen. Weather conditions varied considerably, some- 96 rural sites, regularly distributed inside a times within a trip, more often between trips. radius of 200 km around a central position close Wind directions during sampling were fairly evenly to Wu ¨ rzburg, Germany (49°46∞29◊N, 10°00∞50◊E), distributed, but east or west winds were somewhat were visited for sampling during 16 trips in 3 more frequent than winds from north or south successive summers (June to September in (Fig. 2). 1995–1997). Each trip lasted 4–5 days and covered 2.2. Gas chromatography In the laboratory, adsorbed compounds were transferred by thermodesorption directly to the capillary column of a gas chromatograph (Hewlett-Packard 5890 Series II ), in which two non-polar fused silica capillary columns (Supelco SPB-1, Cat. No. 2-5304, 15 m, 0.53 mm ID, 1.5 mm Fig. 1. The 96 sampling sites surrounding the centre near Wu ¨ rzburg, Germany. The radius of the outer circle is 200 km. Letters A–D indicate diVerently oriented rectan- gular crosses along which the sampling trips were per- formed. Equally symbolized sites were visited ﬁrst along Fig. 2. Mean directions and speeds of the wind measured one axis and then along the other. Deviations from a during sampling. Crosses show data obtained at the regular distribution were caused by topographical condi- various peripheral sites, circles show those obtained at tions (open rural areas at higher altitudes rather than in the centre. The directions given for 0 km/h (calm) are valleys were preferred ). meaningless. Tellus 52B (2000), 4 1142 h. g. wallraff and m. o. andreae ﬁlm thickness and Cat. No. 2-4037, 60 m, 0.25 mm taining up to about 350 peaks. The compounds ID, 1.0 mm ﬁlm thickness) were mounted in series. were labelled using a ‘‘C index’’ giving their posi- Desorption was made by heating a Carbotrap tion on the time axis in relation to the reten- tube in 3 steps from 280° to 380°C over a period tion times of n-alkanes (example: C6.5= retention of about 20 min. During this time, helium carrier time halfway between n-hexane and n-heptane). gas was driven through tube and columns using Detected peaks ranged from C3.5 to C20. From a head pressure of 49 psi while the oven was about C10 onwards, however, most of the peaks cooled down to −60°C. Thereafter, starting with were quite small and hence detection, identiﬁca- initial settings of +5°C and 16.2 psi, a slow tem- tion and quantiﬁcation less reliable than in the perature/pressure programme was run with a con- ﬁrst part of each chromatogram. stant ﬂow of approx. 1 ml/min over 420 min, which Chromatograms obtained from diVerent sites at ended at 240°C. Peaks were detected and quanti- diVerent times varied considerably with respect to ﬁed by means of a ﬂame ionization detector (FID). numbers and sizes of peaks. Nevertheless, a typical Since ratios between peak areas inside each chro- pattern of peaks was observed very consistently. matogram were the values of interest, no calib- This pattern was particularly clear in the ﬁrst part rations were necessary. Small variations in of each trace between C3.5 and C8. In this range retention times of diVerent chromatograms were of retention times we found all 16 peaks that could semi-automatically equalized, peak areas integ- be clearly identiﬁed in each of the 224 chromato- rated, and compounds indexed according to their grams (Fig. 3). The following analysis deals mainly standardized retention times. with these 16 compounds, because they show most Each sample provided a chromatogram con- clearly the fundamental results of this study. Fig. 3. Parts of 2 representative chromatograms showing the 16 always and everywhere observed peaks. Numbers indicate their C indices (in bold italics with preﬁx ‘‘C’’: 6 ‘‘top compounds’’ according to Fig. 4B). Names of com- pounds identiﬁed by application of the pure substance are given in the bottom graph. Ordinates are in relative but corresponding units; note the diVerent ranges. Inside the boxes are example calculations giving areas (relative units) and resulting ratios for one pair of substances (see text). Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1143 2.3. Statistical algorithms versus A, accordingly, to a ± 180°). Thus, e as a measure of eccentricity indicates a possible spatial The basic unit used for statistical analyses was gradient with an increasing ratio of A versus B in the amount of one particular substance as a the direction a . percentage of the sum of two or more substances N In some calculations, we considered only the within the same chromatogram (see examples C4.7 directions of sampling sites from the centre, not versus C5.3 in the boxes of Fig. 3). Pairwise coup- their distances. In these cases we computed x= ling of the 16 omnipresent peaks resulted in 120 sin b and y= cos b, where b is the direction of the possible combinations. The percentage of either respective site from the centre (clockwise from peak was calculated for each such combination in north). In analogous calculations, b represented each chromatogram. The resulting percent values the direction of wind during sampling. Data were per compound were then standardized in terms of tested for statistical signiﬁcance using Monte diVerences from the mean of a sample of chromato- Carlo methods as described in Section 3. grams divided by the standard deviation (s.d.). Thereby, each compound was represented with equal weight, independently of its original percent- 3. Results age. Standardized diVerences, d, are expressed in a scale of ±s.d. units per sample, where ‘‘sample’’ 3.1. Statistical evidence for large-scale atmospheric may mean either the whole set of chromatograms gradients obtained at the peripheral sites (n= 192), the 12 3.1.1. T he sample/site mixing approach. With a symmetrical sites visited per trip, or any other limited number of sites and measurements, com- selected part of the whole set. plete uniformity cannot be expected even if there In a next step, the diVerences d were associated is no general bias in irregularly scattered standard- with the spatial distribution of their sites of origin. ized diVerences, d, from the overall mean of a ratio In relation to the centre of the ﬁeld and to north, between given compounds. Even on a pure chance x and y values for each site were determined ( km) basis, the value for eccentricity, e, always deviates and multiplied by the d value obtained for this from zero to some degree. Expectations on this site on the respective day. From W xd and W yd basis can be calculated if the d values are linked (resulting from n chromatograms) a mean vector to the 96 sites at random (the two chromatograms was calculated (Mardia, 1972) with gained at the same actual site were always linked length or eccentricity to the same random site). The cross-hatched histo- gram in Fig. 4A shows the average frequency 2 2 n n distribution of e for the 120 pairs of 16 omnipres- ∑ x d + ∑ y d S i i i i A B A B ent substances resulting from 105 independent i=1 i=1 e= (km) (1) random replications. The ﬁlled columns show how the distribution obtained from the actual linkage and direction between data and sites of their origin deviates from this expectation. The actual distribution is ∑ x d i i far beyond the range that could ever be expected a =arc tan i=1 (2) N on the basis of pure chance. The maximum degree ∑ y d of geographical sorting of positive and negative i i i=1 values of d, as shown in the insert of Fig. 4A, was (degrees clockwise from north). not achieved in any one of the 105 random Since sites were symmetrically distributed and permutations. average d is, per deﬁnition, zero, the resulting The circular insert in Fig. 4A (radius 200 km) vector should also be zero if the deviations from shows the relative portions per site of the highest the average proportion between the two sub- ranking pair of compounds, C4.0 compared with stances were uniformly distributed over the sites. C5.3. Filled circles indicate sites with C4.0 above The more e deviates from zero, the more the ratio and C5.3 below its average per trip, empty circles of substance A (with respect to B) deviates from those where the relation is reversed; sizes of circles its overall mean towards the direction a (and B are proportional to absolute standardized diVer- Tellus 52B (2000), 4 1144 h. g. wallraff and m. o. andreae Fig. 4. Geographical eccentricity (e) of means of 16 compounds pairwise compared with each other. A: Frequency distribution of eccentricities of the 120 pairs of substances. Filled columns give the actually observed distribution, cross-hatched columns the average distribution of 105 replications attributing the 192 chromatograms to the 96 sites at random. Thin curves give averages of the 10 (=0.01%) random replications that reached the highest medians (solid line), the best correlations with the original distribution ( broken line) and the highest mean for the uppermost 5 values per run (dotted line). The ﬁlled dot at 16.5 km symbolizes the median of the actually observed distribution, the dot at 7.7 km the overall random median with the 99.9% range of individual medians. The circular insert shows the spatial distribution of ratios of the highest ranking pair of compounds (see text). B: The 16 substances ranked according to the mean eccentricity resulting from pairwise comparison with each of the other 15 substances. Actual values are shown by the uppermost thick curve. The other curves give chance expectations as resulting from 105 replications in which chromatograms were attributed to sites at random. In each run, separate ranking was made according to mean eccentricity (independently of the substance order). The overall random median is shown by the lowermost curve, the upper limits for 95, 99 and 99.9% of the random runs by the three thinner curves. Open unconnected circles indicate lengths of the second-order mean vectors calculated from every 15 e values under consideration of direction a (if all 15 a values were identical, results were identical with the ﬁlled circles). The N N associated second-order mean direction is given along the abscissa together with the C index of the respective substance. Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1145 ences d. Ellipses indicate means and 95% conﬁd- (see horizontal lines 99% and 95% in Fig. 4B) ence areas of the ‘‘black’’ and the ‘‘white’’ sub- and down to rank 9 in less than 8%. Thereafter, sample (one mirrored to the other would result in these chance probabilities increase from 10% to the common eccentricity e= 37.9 km, direction 28% in ranks 10–13 and ﬁnally to 79–83% in a = 309° for C4.0 and 129° for C5.3 ). ranks 14–16. The sequence in these lower ranks The shape of the black histogram in Fig. 4A, is, therefore, uncertain, but identiﬁcation of the reﬂecting the actually observed frequencies, sug- highest-ranking substances should be quite gests a continuous transition from irregularity to reliable. spatial distributions showing a smaller or larger The computational methods we used could be degree of regular directionality. The top values modiﬁed in many details without substantially are not exceptional cases but rather mark the aVecting the principal outcome. Besides the ‘mean upper end of a fairly normal distribution. If vector approach’, as described above, we applied random samples were selected to contain particu- a ‘rotating axis approach’, with similar results. In larly high values, the mass of values remained in this case, an axis running through the centre is the typical random range (Fig. 4A, dotted curve). rotated stepwise through 180°, and in each posi- The real data indicate, in contrast, a general trend tion the cosine of each site projected on the axis towards a larger-scale spatial order in the (i.e., the distance along the axis) acts as x in a atmosphere. regression analysis, while the d values act as y. It is important, but not suYcient, to know that The axis obtaining the highest correlation coeY- some ratios between substances vary with some cient is very similar to the above a , and the directional regularity. Fig. 4B gives, in addition, distribution of maximum correlation coeYcients an individual ranking of the 16 compounds is similar to that of e (although in a diVerent together with the direction of the average vector scale). If the whole sample of 192 chromatograms calculated for each of them. The non-connected is split into two parts (e.g., according to months open symbols in this graph indicate that some of or time of day, according to the two rounds with the compounds (ranks 1, 2, 3, 5, 13) show very visits of the same 96 sites, or sites alternating so consistent ( but diVerent) unidirectional vectors, that no site is included in both sub-samples), the while others (ranks 4, 9, 11) are quite variable in results diVer in details, statistical signiﬁcances tend their directedness depending on the substances (naturally) to be lower, but the overall outcome with which they are compared. These latter com- remains unaVected. pounds constitute a fairly neutral background against which the spatially variable compounds 3.1.2. T he error adding approach. A second, vary in their proportional ratio. completely diVerent approach to testing the reality Such conclusions, however, should not be of gradients is based on the consideration that, in extended, without reservation, to all 16 com- the case of pure random eVects, other proportional pounds we investigated in detail. Although the distributions of peaks within each chromatogram data set as a whole is incompatible with a pure should also produce similar levels of eccentricity. random origin, irregular variability may be inter- To test this possibility, a random error value in mingled with fairly regular trends. We must expect the range of ±s was added to each peak-area that the mean eccentricity levels of the individual value, where s is the standard deviation of a compounds are co-determined by chance. normal distribution in percent of the peak area. Although, for instance, the chance probability is Thus, with increasing s, original peak sizes were less than 0.1% that a compound ranking at posi- increasingly modiﬁed at random to either smaller tion 13 has a mean e value of 14.1 km, this value or larger values. Fig. 5A shows that, along with is reached or exceeded by the compound ranking this increase of s, the average eccentricity of the at position 1 in 28% of the random runs and in uppermost 10% of the 120 pairs of substances some rare cases even by the rank 2 or 3 com- decreased. Beyond of s=±40%, when the evalu- pounds. The probability of chance ranking is quite ated peak areas reﬂect only little of their original low, however, in the highest ranks. The e level of proportions, the original level of eccentricity e was ranks 1–3 has been reached in less than 1% of never reached by any random set of manipulated the random runs, down to rank 5 in less than 5% data. With smaller ranges of superimposed error, Tellus 52B (2000), 4 1146 h. g. wallraff and m. o. andreae Fig. 5. Mean eccentricity of the 12 highest ranking pairs of substances, out of 120 pairs, with random errors superim- posed on the individual peak areas in the 192 chromatograms. A: The heavy curve gives the original (s=0 ) and the mean of 104 random replications with increasing ranges of random errors superimposed on the peak areas. Ranges for 95, 99 and 99.9% of each of the 11 sets of 104 random runs are drawn by thin lines. B: Mean diVerences in average eccentricity between all 16 substances modiﬁed by superimposed random errors (s=±50%; result per run set to zero) and one substance (C index on the abscissa) left in its original state in else identical random runs. 100 replications per selected substance; means and conﬁdence intervals 95, 99 and 99.9%. slightly increased eccentricities were achieved in a repeated with s=±50%, but the peaks of only small percentage of runs, but the average started 15 compounds were modiﬁed by an artiﬁcial to decrease already with about s=±5%, which random error, whereas the peaks of one compound may be the approximate error range of actual were left at their original sizes. In identically measurement. randomized runs, one with 16 and one with only This approach also allows us to rank the indi- 15 compounds manipulated, the diVerence De vidual compounds (Fig. 5B). Calculations were between average e of the uppermost 12 of 120 Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1147 peak pairs was determined. The results of such 3.2. Multi-compound gradient ﬁelds double runs, with each compound successively left Considering not absolute values but ratios, pair- in its original state, revealed a ranking of sub- wise combination of only 2 compounds, resulting stances that is identical in its ﬁrst 6 positions with in one ratio per site, would be insuYcient for those found with the ﬁrst procedure (compare position ﬁxing in a 2-dimensional plane. A min- Fig. 4B). For these 6 substances, the diVerence imum of 3 compounds would be necessary, whose from pure random results in Fig. 5B is so large, ratios entail two degrees of freedom, but with that a general chance basis can be excluded. noisy signals a larger number of gradients would lead to more reliable conclusions (WallraV, 1989a). 3.1.3. A look at not always recorded substances. Combining the 16 substances in 560 diVerent The 6 top substances, selected from a total of triplets yields similarly large diVerences of spatial more than 300 observed peaks by the above regularities from random expectations as the pair- algorithms, demonstrate the existence of diVerently wise combinations shown in Fig. 4. The same is oriented spatial atmospheric gradients, but do not true for higher-order combinations of up to 6 necessarily represent the real top group of sub- compounds. stances with the strongest spatial gradients. In When the 6 highest-ranking substances (Figs. order to avoid any possible manipulation of data, 4B and 5B) are combined, the relative portion of peak integration in the chromatograms followed at least ﬁve of them shows an overall spatial an automated procedure throughout without indi- gradient which is far from simple monotony but vidual optimization. Consequently, quantiﬁcation clearly diVerent in its direction among the sub- and even identiﬁcation of smaller peaks, particu- stances (Fig. 6). Various other combinations (trip- larly when superimposing each other, was often lets, quadruplets, etc.) lead to basically similar associated with considerable ranges of possible ‘chemical landscapes’. It should be noticed that error. Since absolute concentrations of all com- these patterns represent neither snapshots of a pounds varied extremely, the less abundant com- momentary state nor long-term averages, but pounds were not detected in the more dilute rather result from a mixture of many spot checks chromatograms at all, but may, nevertheless, have obtained under quite variable weather conditions been present at very low concentrations. When and wind directions. These conditions were fairly the threshold for inclusion of a substance in the evenly distributed over the whole area. Due to the analysis was set as being observed in at least 80% symmetry of each trip, the observed spatial pat- of the chromatograms, 72 substances met this terns are unlikely to have been caused by hidden criterion. The highest ranking 18 compounds temporal or condition-dependent biases. (25%), in descending order, were then the follow- ing (those already included in the set of 16 are in brackets; e values of the ﬁrst 8 are particularly 3.3. EVects of wind direction and origin of air outstanding): C7.5, C6.9, (C5.1), (C5.3), (C4.0), masses C17.3, C17.6, C14.8 — (C4.7), C5.58, (C6.6), C13.2, 3.3.1. L ittle inﬂuence on directions of spatial C11.5, C18.5, C7.2, C18.0, C19.9, (C3.5). This gradients. When considering potential spatial gra- collection suggests that spatial proportional gradi- dients of trace gases, winds deserve particular ents do not only occur in the lighter compounds attention in a ‘negative’ as well as in a ‘positive’ below C8, as considered in this study in detail, way. The ‘negative’ aspect is quite obvious and but also in substances with a considerably large the main reason why navigation based on airborne molecular weight. As quantities of these substances odours is often presumed unfeasible. In many were always quite small, however, they could not regions, and particularly in those in which homing be recorded and evaluated reliably by the methods of birds has been shown to occur, winds are we applied ( but might be detected and evaluated extremely variable. The atmosphere covering a by birds). It seems very likely that the present given area is rapidly exchanged by air masses investigation, due to its methodological limita- arriving from diVerent directions over diVerent tions, underestimates the frequency and level of types of land or sea. Intuitively and with some spatial regularities in the chemical composition of good will, one might consider it possible that a the atmosphere. certain pattern of gradients is fairly consistently Tellus 52B (2000), 4 1148 h. g. wallraff and m. o. andreae Fig. 6. Proportional relationships between the six compounds ranking highest in Figs 4B and 5B. In each of the 192 chromatograms the peak-area sum of all 6 substances was set to 100 and percentages of the individual compounds determined. For each substance, diVerences from the mean percentage per trip were calculated and transformed into a scale of standard deviation. From the resulting two-dimensional point diagrams (analogous to the insert in Fig. 4A), complete ‘‘landscaspes’’ were computed by interpolation. Areas with values above average are bright, those with values below average are dark. Small diagrams show means± s.e.m. parallel to the optimal axis whose direction is indicated. The abscissa gives the distance from the line running perpendicularly to this axis through the centre; the ordinate gives the diVerence from the overall mean per substance in terms of standard deviation. Distance classes for averaging are 0–60 km, 60–120 km, and 120–200 km. restored each time when winds are blowing, for to the gradients of proportions between our six instance, from the east. With westerly winds, how- top substances selected according to their spatial ever, one would expect a quite diVerent pattern. regularities (Fig. 7). If the sample of 167 chromato- This expectation has not been met with respect grams collected under conditions of wind speed Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1149 Fig. 7. Mean percentages of the 6 top substances ( Fig. 6 ) in various directions from the centre with varying directions of the wind during air sampling (wind speed 5km/h). Directions of sites from the centre (irrespective of distance) were combined in angular classes and running means of the percentage per compound determined in steps of 5° around the circle (unequal angular distribution of sites thereby compensated by averaging). Curves were calculated for two pairs of mutually exclusive sub-sets of chromatograms: with winds from semicircles north and south (thick curves without dots) and semicircles west and east (thin curves with dots). Numbers of included chromatograms are given in brackets. Site-independent means per wind class are shown by thin horizontal lines. Original, non-standard- ized percentages (ordinates) indicate the ranges of variability for each substance. of at least 5 km/h was subdivided into two sub- tion of sites in a roughly similar way in both sub- samples obtained with winds from the eastern or samples (Fig. 7: two dotted curves per diagram). the western semicircle, respectively, the propor- The same was true, if winds from the northern tions varied in dependence on the spatial distribu- and the southern semicircle were compared (two Tellus 52B (2000), 4 1150 h. g. wallraff and m. o. andreae Fig. 8. Mean percentages of the 6 top substances (Fig. 6 ) in various directions from the centre with air masses of diVerent origin as indicated. Graphs are analogous to those in Fig. 7. thick curves). In each substance, the proportional maritime air masses (types mT, mTp, mPt, mP, maximum and minimum was located in similar mPa) and continental air masses (cP, cPt, cTp), spatial directions independently of whether the respectively, are fairly similar to those obtained wind came preferentially from north, south, west with westerly and easterly winds, respectively or east. (Fig. 8). This similarity is largely based on the fact Analogous subdivisions were made with respect that in 77% of the cases winds from the western to the type of air mass as categorized in the semicircle were correlated with maritime air weather reports for each day (Scherhag, 1948), masses and winds from east with continental air with similar results. Curves obtained for days with masses. No such correlation existed between winds Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1151 from north and ‘polar’ air masses (cP, cPt, 192 chromatograms was reduced to 167. On the mPt, mP, mPa) and between winds from south other hand, the 32 chromatograms obtained from and ‘tropical’ air masses (cTp, mT, mTp). collection at the centre ( 22 with wind speed Nevertheless, the curves calculated for these two 5km/h) could be included, so that a total of categories of air masses also roughly ﬁt into the 189 chromatograms could be analyzed with general patterns that are typical for the diVerent respect to a possible eVect of wind direction compounds (Fig. 8). (Fig. 2). As Fig. 9A shows, such an eVect appears The fairly uniform shapes of the curves for each to exist. The distribution of the resulting vector compound in Figs. 7 and 8 indicate that the lengths diVers markedly from random expectation; directions of the gradients were not very much chance probability for the longer vectors to occur modiﬁed by varying wind conditions and air is p< 0.001. However, the top values did not masses. This does not mean, however, that winds wholly reach the level found in analogous calcula- and air masses had no inﬂuence at all. The average tions of spatial vectors (p< 0.0001). portion of each compound in the whole area, but Fig. 9B shows the results for the individual not its directional slope in relation to the other compounds. The data base is less complete than compounds, did vary with wind direction and with in Fig. 9A, but space-dependent and wind-depend- air mass type (see horizontal lines in Figs. 7 and 8). ent values are more immediately comparable, as both calculations comprise the same 167 data sets 3.3.2. Inﬂuence on the level of ratios between obtained at the peripheral sites with wind speeds compounds. This latter facet leads to the potential 5km/h. Both aspects interfere and ‘‘compete’’ ‘positive’ aspect of winds, i.e., to their assumed with each other: EVects of winds are intermingled functional involvement in bird navigation. In with spatial eVects and vice versa. Fig. 9B suggests order to derive, at an unfamiliar remote site, that spatial regularities are more clearly expressed, information on the current position relative to and statistically better signiﬁcant, than wind-con- home from proportions between compounds, birds nected regularities. The ordering of substances need to know the direction of the gradient of each listed along the abscissa in Fig. 9B shows that of the compounds of which they make use. As there is, by and large, a negative correlation mentioned above, homing experiments suggest the between space dependence and wind dependence hypothesis that pigeons are able to deduce these of the individual substances. Five of the six com- directions, without leaving the home site, from pounds composing the top group under spatial changes of compound ratios that are correlated aspects (Figs. 4B, 5B) are assembled at the lower with changes of wind direction. Thus, it is neces- end when ranked according to their regular sary to investigate whether such correlations do dependence on wind direction. Thus, the sub- exist, and if they do, which relationships they stances showing the clearest spatial gradients in show to the spatial gradients of proportions. their proportional presence appear less aVected Wind directions were used in our calculations by varying winds. in essentially the same way as described for the spatial directions. Instead of the compass direction 3.4. Relationships between spatial gradients and of each site from the centre, the direction of wind wind-related variations measured during air sampling was used to com- pute mean vectors. Distance from the centre could Finally, we look for directional relationships have been replaced by wind speed, but was not, between spatial gradients and shifts of ratios because possible correlations were expected much between substances depending on wind direction. more with directions rather than with the more For this purpose, we mimic the situation of pigeons variable speeds of winds during the short period whose loft is in the centre of our sampling ﬁeld. of measurement (and in fact, results were less clear During their long stay there, and before any when speeds were additionally included ). displacement, birds are assumed to associate the However, a threshold of minimum speed was set proportional composition of a set of substances at 5 km/h, because with lower speeds also direc- with current wind directions. Unlike the pigeons, tions were unreliable or could not be determined we do not have continuous long-term measure- at all. With this threshold, the original sample of ments ‘at home’ but have only a limited number Tellus 52B (2000), 4 1152 h. g. wallraff and m. o. andreae Fig. 9. Wind-dependent eccentricity of means of 16 compounds pairwise compared with each other. Graphs ana- logous to Fig. 4. Histograms and solid lines refer to wind data, broken lines (for comparison) to analogously computed spatial data. Only directions were considered, not speeds of wind and distances of sites from the centre (scale for vector length is the unit circle with a maximum of 1 ). Random expectations were calculated from 104 replications. Thin lines in A give means of the 10 (=0.1%) best random runs. Numbers along the abscissa in B are C indices in rank-order sequence for winds (above) and sites ( below); the six compounds ranking highest in Figs. 4B and 5B are in italics. Corresponding mean directions (degrees clockwise from north) for winds (above) and sites ( below) along the top of B. For further explanations see Fig. 4; for selection of data see text. of spot checks. Nevertheless, our data imply a pairs of substances, for triplets or for larger groups, general systematic relationship between the direc- for the 16 omnipresent compounds or for larger tions of spatial gradients and the directions of sets. The result is, in principle, always the same. maximum changes of ratios depending on winds. Fig. 10A shows the angular relationships derived Both kinds of directions can be calculated for from the 2556 pairwise combinations that are Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1153 Fig. 10. Angular relationships between corresponding directions of wind-dependent and site-dependent eccentricities of compound ratios. A: Results obtained from pairwise combinations of 72 substances. Cross-hatched columns and curves show deviations of all directions of average increase depending on wind from the corresponding spatial gradient (n= 2556 pairs of compounds). Filled columns include only the more distinct eccentricities (e> 0.075 for both space and wind; n=525 pairs). Histograms refer to all wind data (5km/h; n=189 samples), curves only to winds at the peripheral sites (solid line; n= 167 samples) or at the central site ( broken line; n=22 samples), respect- ively. Spatial data include always the total of 192 samples collected at the peripheral sites. B: Deviations of ‘wind slope’ from ‘spatial slope’ of the 16 individual omnipresent compounds, each paired with its 15 partners. Triangles give mean vectors, each calculated from 15 directions (ordinate 0= uniformly distributed, 1=all identical ). The 6 top substances ( Fig. 6 ) are marked by their C indices and by ﬁlled triangles. possible with the 72 compounds detected in at (data sample for centre quite small ) or eVects of least 80% of the air samples. In most cases, the local conditions at the central site. steepest increase in the ratio of a substance in For most compounds, the wind mean deviates relation to wind direction occurs in a direction clockwise from the spatial mean, but the degree deviating clockwise from its steepest spatial gradi- of divergence is quite variable (Fig. 10B). C5.3, for ent. The mean deviation is about 50°. The fairly instance, one of the ‘‘top substances’’, shows an good agreement between wind-related directions angle diVering only slightly from zero. The propor- calculated from samples taken at the central site tion of this substance tends to decrease geograph- and at the peripheral sites (curves in Fig. 10A) ically from about SE to NW (Fig. 4), and makes it highly unlikely that the result is due to correspondingly, it tends to be highest with winds chance. The slight diVerence between the two from SE and lowest with winds from NW. In curves may indicate either minor eVects of chance contrast, C4.0, another ‘‘top substance’’, spatially Tellus 52B (2000), 4 1154 h. g. wallraff and m. o. andreae increasing from SE towards NW (Fig. 4), showed ized regions of the globe. The most important its highest ratio not with winds from NW, but sources are transport related; tailpipe emissions, with winds from NE. gasoline evaporation and production account for These results suggest that there is a systematic over half the VOCs found in urban and rural directional relationship between relative amounts regions in North America and Europe (Lin and of substances as correlated with geographical posi- Milford, 1994; Clarke and Ko, 1996; ScheV et al., tion and with wind direction. According to the 1996). In Germany, anthropogenic emissions of available data, this relationship is not so simple VOCs (not including methane) are thought to that, for instance, with northerly winds a relative exceed biogenic ones by a factor of 2–10 (Guenther increase of those substances occurs whose relative et al., 1995; Simpson et al., 1995). The substances portion is spatially larger in the north than in the that in our study showed strong spatial or wind- south. Instead, northerly winds seem to cause a related structure can be related to a variety of relative increase of substances that are spatially emission sources: The butanes (C3.5, C4.0) are most abundant in the northwest or even in the emitted during natural gas handling, gasoline west. The generally observed clockwise shift of evaporation and reﬁning. The pentanes (C4.7), the ‘‘wind’’ against ‘‘space’’ might be a consequence of hexanes (C5.3, C5.69, C5.8), benzene (C6.6) and the fact that air masses usually follow curved toluene (C7.7) are characteristic of automobile trajectories. In an anticyclonic system, for instance, exhaust. Toluene is emitted also during the use of northerly winds are connected with air masses a variety of coatings (paints). Methyl chloroform arriving from northwesterly areas and hence might (C6.5) is widely used as a degreasing solvent, increase the relative portion of ‘‘northwesterly’’ although its use is now declining since it was substances. Retrospective counts revealed that our banned under the Montreal Protocol. data samples were not collected under equilibrated Due to the uneven geographic distribution of synoptic weather conditions according to Hess source types, the emission ﬁelds of the diVerent and Brezowsky (1977). As far as the days could VOCs in Europe show distinct gradients at scales be adequately categorized under this aspect, 80% of hundreds to thousands of kilometres (B. of the 164 samples we obtained at the peripheral Langmann, pers. comm., 1998). Consequently, air sites were taken during anticyclonic conditions masses arriving from the west and northwest and only 20% in cyclonic conditions. Among the would have passed over long fetches with high 19 usable central-point measurements 84% were emission densities for many of the vehicle-related anticyclonic and 16% cyclonic. In contrast, the VOCs, whereas air masses from easterly directions percentages for the full periods June–September would have been exposed more to emissions from 1995–1997 diVered much less: 58% anticyclonic coal burning. The emissions of biogenic VOCs, and 42% cyclonic. So we cannot decide whether such as isoprene and the monoterpenes, also show the directional relationships shown in Fig. 10 are pronounced geographic gradients. Isoprene emis- generally representative or valid only for certain sions increase by almost an order of magnitude weather conditions. The few data available for from the centre of the study area to a region some cyclonic conditions result in less clearly expressed, 300 km to the WSW, where Europe’s highest but even larger, counterclockwise deviations. So isoprene emissions are occurring (Simpson et al., far, therefore, it cannot be excluded that long- 1995). Because of technical limitations, isoprene term continuous observation might produce an was the only biogenic substance for which we average angular divergence between ‘‘wind’’ and were able to observe signiﬁcant spatial gradients. ‘‘space’’ not very far from zero. However, given the vast diversity of biogenic volatile substances emitted by diVerent species of plants, animals, and microbes (Isidorov et al., 3.5. Chemical identity of the atmospheric tracers 1985), and the close linkage of these emissions to The most prominent substances in our chroma- taxonomic and ecological parameters (Monson tograms are, with the exception of isoprene (C5.1), et al., 1995), it is highly plausible that similar all of predominantly anthropogenic origin. This large-scale gradients exist in other biogenic VOCs situation reﬂects the overwhelming inﬂuence of as well. It has been shown, for example, that the anthropogenic VOC emissions in the industrial- highest monoterpene emissions are in northern Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1155 Europe, where coniferous trees are most abundant accomplished a ﬁrst step from pure speculation to (Guenther et al., 1995). empirical research. Based on the empirical data The inhomogeneous spatial distribution of VOC described above, a computer model (WallraV, sources provides one of the reasons why these 1989b) can be applied by which navigation from substances may show spatial gradients in their any peripheral point to the centre is possible relative concentrations. But even in the absence without previous knowledge of proportional com- of diVerential source patterns, gradients in relative positions in the distant areas, provided that the abundance would develop due to the diVerent rough directions of the gradients are known atmospheric lifetimes of the individual species. The (WallraV, 2000 ). The results shown are consistent lifetimes of the alkanes and aromatics, which are with the hypothesis that this necessary knowledge most prominent in our data set, typically fall in can be obtained, at the birds’ home site, from a the range of 1–7 days, while the lifetimes of correlation between wind direction and propor- isoprene and the monoterpenes are of the order tional composition of a set of airborne odorants. of hours to a couple of days. However, even when It is not yet clear whether this correlation can be these substances are oxidized, they persist in the as simple as hitherto presupposed by the model atmosphere in the form of oxidation products, or whether a more complicated mechanism must which may have considerably longer lifetimes. be assumed. Many more data, simultaneously Considering that lifetimes of a day or two at wind collected over an extended grid of sampling sites, speeds around 10–20 km/h correspond to would be required in order to fully understand enfolding distances of about 200 to 1000 km, it is the interrelations between spatial ratio gradients quite plausible that a combination of diVerential and changes of ratios in dependence on wind source distribution and chemical conversion in directions. In this respect, our data set must be the atmosphere creates spatial gradients on the considered preliminary. scales relevant for navigation of birds. The exist- Our study did not attempt to identify airborne ence of analogous source inhomogeneity and substances that pigeons and other birds actually diVerential reactivity for anthropogenic and nat- use for navigation. However, we have determined ural substances, as discussed above, together with that there are compounds in the atmosphere which the fact that we found gradient structures also for theoretically could be used. One of these sub- isoprene, suggest that it is reasonable to generalize stances was identiﬁed as isoprene, a hydrocarbon our ﬁndings to natural VOCs. emitted by plants. The fact that most of the other substances investigated here cannot be identical, due to their anthropogenic origin, with those to 4. Discussion which avian navigation systems have been adapted during evolution, does not preclude the inference The principal aim of this study was to explore that atmospheric conditions make olfactory nav- whether the atmosphere contains trace gases igation basically feasible. Gradual variations in whose proportional distribution yields fairly con- the ratios of compounds are likely to result from sistent gradients covering several hundreds of kilo- a combination of structured source patterns, vary- metres in various directions. Statistical analysis ing speeds of chemical conversion and turbulent has revealed that this question can be answered horizontal and vertical mixing. In spite of advect- in the aYrmative (Figs. 4–6). Directions of spatial ive transport with the prevailing winds, such gradi- gradients are noisy but remarkably robust against ents may persist over large regions because of the changes of weather, especially of winds (Fig. 7) turbulent character of horizontal and vertical and origin of air masses (Fig. 8). Nevertheless, diVusion in the atmosphere. By advective trans- ratios between substances show a dependence on port with the prevailing winds, such gradients may wind direction as well (Fig. 9). A systematic rela- be displaced as a whole while their directions tionship exists between variations in space and roughly persist over large regions (cf. Figs. 7 and variations correlated with the direction of winds 8). In contrast to the nonspeciﬁc method applied (Fig. 10). here, which could not be focused on particular With these ﬁndings, the search for the atmo- substances known in advance, birds are probably spheric basis of avian olfactory navigation has specialized and most sensitive to a few most Tellus 52B (2000), 4 1156 h. g. wallraff and m. o. andreae suitable compounds which, for methodological atmospheric structures exploitable for navigation reasons, could not be included in the above ana- do exist. Science in general, however, should not lysis but might provide more reliable spatial gradi- be satisﬁed. It should feel challenged to ask how ents than those revealed by the crude means the observed spatio-temporal regularities in the proportional composition of airborne trace sub- employed in this ﬁrst approach. stances originate and how they achieve their relat- Regional climatic and geomorphological fac- ive stability. In particular, two avenues should be tors, and factors dependent on them, probably explored: First, the simulation of the atmospheric determine the sizes of suYciently monotonic ﬁelds distributions of speciﬁc VOCs in regional models, of ratio gradients. It is to be expected that on the using known source distribution ﬁelds, should be European continent, with its small-scaled coastal used to obtain gridded data sets which can be and mountainous structures and rapidly changing subjected to the same statistical analysis as applied weather conditions, sizes of gradient ﬁelds are to the ﬁeld data used in our study. This would limited and regionally variable. If, nevertheless, make it possible to investigate which physical and remarkably stable and relatively extended gradient chemical processes are involved in creating the ﬁelds exist even there, one may expect much more structures we observed. Second, observational extended roughly monotonic gradients over and studies involving synchronous sampling over a inside the oceans, where air and water currents grid of sites should be designed based on the are much more regular. Conditions for large-range results from these modeling exercises. These olfactory navigation might be particularly favour- experiments may provide a critical test of how able for marine animals such as albatrosses, pet- well the interactions of advective and turbulent rels, turtles, whales, salmons and other ﬁshes, transport, as well as source and sink processes, which are known to migrate and navigate over are represented in regional atmospheric models. distances of thousands of kilometres over or in the open sea (Quinn and Dittman, 1992; Papi and Luschi, 1996; Lohmann et al., 1999). Olfactory 5. Acknowledgements landscapes over seemingly featureless oceans, of which sea birds apparently make use in smell- We thank G. Schebeske for his help in estab- guided foraging ﬂights (Nevitt et al., 1995; Nevitt, lishing gas chromatographic techniques. We are 1999), could be much more eVectively utilized for grateful to B. Langmann for providing us with goal ﬁnding over long distances, if they would information on the spatial distribution of hydro- include extended ratio gradients. carbon emissions in Europe. Data on air masses Birds as well as biologists investigating bird and synoptic weather conditions were kindly pro- homing can be satisﬁed by the apparent fact that vided by Deutscher Wetterdienst, Munich. REFERENCES Able, K. P. 1996. The debate over olfactory navigation Fujita, E. M., Watson, J. G., Chow, J. C. and Magliano, by homing pigeons. J. Exp. Biol. 199, 121–124. K. L. 1995. Receptor model and emissions inventory Atlas, E. L., Li, S.-M., Standley, L. J. and Hites, R. A. source apportionments of nonmethane organic gases 1993. Natural and anthropogenic organic compounds in California San Joaquin Valley and San Francisco in the global atmosphere. In: Global atmospheric chem- Bay area. Atmos. Environ. 29, 3019–3035. ical change (eds. C. N. Hewitt and W. T. Sturges). Ganzhorn, J. U. and PaVrath, D. 1995. Patterns in air Elsevier Applied Science, London, pp. 313–381. pollution as model for the physical basis for olfactory Becker, J. and Van Raden, H. 1986. Meteorologische navigation in pigeon homing. J. Ornithol. 136, Gesichtspunkte zur olfaktorischen Navigationshypo- 159–165. these. J. Ornithol. 127, 1–8. Gould, J. L. 1998. Sensory bases of navigation. Current Bingman, V. P. 1998. Spatial representations and homing Biology 8, R731–R738. pigeon navigation. In: Spatial representation in animals Guenther, A., Hewitt, C. N., Erickson, D., Fall, R., Geron, (ed. S. Healy). Oxford Univ. Press, Oxford, pp. 69–85. C., Graedel, T., Harley, P., Klinger, L., Lerdau, M., Clarke, A. G. and Ko, Y. H. 1996. The relative signiﬁc- Mckay, W. A., Pierce, T., Scholes, B., Steinbrecher, R., ance of vehicular emissions and other emissions of Tallamraju, R., Taylor, J. and Zimmerman, P. 1995. A volatile organic compounds in the urban area of Leeds, global model of natural volatile organic compound UK. Science of the total environment 190, 401–407. emissions. J. Geophys. Res. 100, 8873–8892. Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1157 Hess, P. and Brezowsky, H. 1977. Katalog der Großwet- Scherhag, R. 1948. Neue Methoden der Wetteranalyse und terlagen Europas. Ber. Dtsch. Wetterdienst. 15, no. Wetterprognose. Springer, Berlin. Schmidt-Koenig, K. and Ganzhorn, J. U. 1991. On the 113, 1–14. problem of bird navigation. Persp. Ethol. 9, 261–283. Hitchcock, H. B. 1952. Airplane observations of homing Simpson, D., Guenther, A., Hewitt, C. N. and Stein- pigeons. Proc. Amer. Phil. Soc. 96, 272–289. brecher, R. 1995. Biogenic emissions in Europe (1 ). Isidorov, V. A., Zenkevich, I. G. and IoVe, B. V. 1985. Estimates and uncertainties. J. Geophys. Res. 100, Volatile organic compounds in the atmosphere of for- 22,875–22,890. ests. Atmos. Environ. 19, 1–8. Singh, H. B. and Zimmerman, P. B. 1992. Atmospheric Lin, C. and Milford, J. B. 1994. Decay-adjusted chemical distribution and sources of nonmethane hydrocar- mass balance receptor modeling for volatile organic bons. In: Gaseous pollutants: characterization and cyc- compounds. Atmos. Environ. 28, 3261–3276. ling (ed. J. O. Nriagu). John Wiley, New York, Lohmann, K. J., Hester, J. T. and Lohmann, C. M.F. pp. 177–235. 1999. Long-distance navigation in sea turtles. Ethol. Wagner, G. 1970. Verfolgung von Brieftauben im Heli- Ecol. Evol. 11, 1–23. kopter. Rev. Suisse Zool. 77, 39–60. Mardia, K. V. 1972. Statistics of directional data. Aca- Waldvogel, J. A. 1987. Olfactory navigation in homing demic Press, London. pigeons: are the current models atmospherically real- Monson, R. K., Lerdau, M. T., Sharkey, T. D., Schimel, istic? Auk 104, 369–379. D. S. and Fall, R. 1995. Biological aspects of con- Waldvogel, J. A. 1989. Olfactory orientation in birds. structing volatile organic compound emission invent- Current Ornithology 6, 269–321. ories. Atmos. Environ. 29, 2989–3002. WallraV, H. G. 1989a. Simulated navigation based on Mukund, R., Kelly, T. J. and Spicer, C. W. 1996. Source unreliable sources of information (models on pigeon attribution of ambient air toxic and other VOCs in homing, part 1 ). J. T heor. Biol. 137, 1–19. Columbus, Ohio. Atmos. Environ. 30, 3457–3470. WallraV, H. G. 1989b. Simulated navigation based on Nevitt, G. 1999. Foraging by seabirds on an olfactory assumed gradients of atmospheric trace gases (models on pigeon homing, part 2 ). J. T heor. Biol. 138, landscape. Am. Scientist 87, 46–53. Nevitt, G., Veit, R. R. and Kareiva, P. 1995. Dimethyl 511–528. WallraV, H. G. 1990a. Navigation by homing pigeons. sulﬁde as a foraging cue for Antarctic procellariiform Ethol. Ecol. Evol. 2, 81–115. seabirds. Nature 376, 680–682. WallraV, H. G. 1990b. Long-distance navigation of Papi, F. 1989. Pigeons use olfactory cues to navigate. homing pigeons based on airborne olfactory signals. Ethol. Ecol. Evol. 1, 219–231. In: ISOT X: Proc. 10th Int. Symp. Olfaction and taste Papi, F. 1991. Olfactory navigation. In: Orientation in (ed. K. B. Døving). GCS A/S,Oslo, pp. 26–35. birds (ed. P. Berthold). Birkhauser, Basel, pp. 52–85. WallraV, H. G. 1991. Conceptual approaches to avian Papi, F. and Luschi, P. 1996. Pinpointing ‘‘Isla Meta’’: navigation systems. In: Orientation in birds (ed. P. Ber- the case of sea turtles and albatrosses. J. Exp. Biol. thold). Birkha ¨user, Basel, pp. 128–165. 199, 65–71. WallraV, H. G. 1996. Seven theses on pigeon homing Piccot, S. D., Watson, J. J. and Jones, J. W. 1992. A deduced from empirical ﬁndings. J. Exp. Biol. 199, global inventory of volatile organic compound emis- 105–111. sions from anthropogenic sources. J. Geophys. Res. 97, WallraV, H. G. 2000. Simulated navigation based on 9897–9912. observed gradients of atmospheric trace gases (models Quinn, T. P. and Dittman, A. H. 1992. Fishes. In: Animal on pigeon homing, part 3 ). J. T heor. Biol., in press. homing (ed. F. Papi). Chapman & Hall, London, WallraV, H. G., Kiepenheuer, J., Neumann, M. F. and pp. 145–211. Streng, A. 1995. Homing experiments with starlings ScheV, P. A., Wadden, R. A., Kenski, D. M., Chung, J. deprived of the sense of smell. Condor 97, 20–26. and WolV, G. 1996. Receptor model evaluation of the Wiltschko, R. 1996. The function of olfactory input in southeast Michigan ozone study ambient NMOC pigeon orientation: does it provide navigational measurements. J. Air & Waste Manag. Ass. 46, information or play another role? J. Exp. Biol. 199, 1048–1057. 113–119. Tellus 52B (2000), 4 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Tellus B: Chemical and Physical Meteorology Taylor & Francis http://www.deepdyve.com/lp/taylor-francis/spatial-gradients-in-ratios-of-atmospheric-trace-gases-a-study-0ZqEZqwu4x

Loading next page...

References (29)

(1996)
Seven theses on pigeon homing Piccot
Olfaction and taste Papi, F. 1991. Olfactory navigation
(1996)
The debate over olfactory navigation by homing pigeons
, 199
R. Scherhag (1948)
Neue Methoden der Wetteranalyse und Wetterprognose
(1990)
1990b. Long-distance navigation
(1998)
Sensory bases of navigation. Current these
(1991)
Conceptual approaches to avian
(1985)
Volatile organic compounds in the atmosphere of for-22,875-22,890. ests
(1986)
Meteorologische Gesichtspunkte zur olfaktorischen Navigationshypo- 159–165
(1952)
Airplane observations of homing
(1992)
Decay-adjusted chemical distribution and sources of nonmethane hydrocarmass balance receptor modeling for volatile organic bons
K. Mardia (1972)
Statistics of Directional Data
(1995)
Homing experiments with starlings Scheff, P
(1992)
Simulated navigation based on 9897-9912. observed gradients of atmospheric trace gases
Able K. P (1996)
121
J. Exp. Biol, 199
(1989)
Foraging by seabirds on an olfactory assumed gradients of atmospheric trace gases (models on pigeon homing, part 2)
(1996)
Pinpointing ‘‘Isla Meta’’: navigation systems
(1995)
Olfactory orientation in birds. structing volatile organic compound emission invent
(1990)
Long-distance navigation of Papi, F. 1989. Pigeons use olfactory cues to navigate. homing pigeons based on airborne olfactory signals
(1996)
The debate over olfactory navigation Fujita, E
F. Papi (1989)
Pigeons use olfactory cues to navigate
Ethology Ecology & Evolution, 1
(1995)
Patterns in air ical change
K. Lohmann, J. Hester, Catherine Lohmann (1999)
Long-distance navigation in sea turtles
Ethology Ecology & Evolution, 11
(1989)
Source unreliable sources of information (models on pigeon attribution of ambient air toxic and other VOCs in homing
(1985)
Estimates and uncertainties
(1995)
Receptor model and emissions inventory by homing pigeons
(1972)
Olfactory navigation in homing demic Press, London. pigeons: are the current models atmospherically real-Monson
(1995)
Dimethyl Wallraff, H
A. Clarke, Yoneda Ko (1996)
The relative significance of vehicular emissions and other emissions of volatile organic compounds in the urban area of Leeds, UK
Science of The Total Environment

Publisher: Taylor & Francis
Copyright: © 2000 The Author(s). Published by Taylor & Francis.
ISSN: 1600-0889
DOI: 10.3402/tellusb.v52i4.17084
Publisher site: See Article on Publisher Site

Abstract

T ellus (2000), 52B, 1138–1157 Copyright © Munksgaard, 2000 Printed in UK. All rights reserved TELLUS ISSN 0280–6509 Spatial gradients in ratios of atmospheric trace gases: a study stimulated by experiments on bird navigation By HANS G. WALLRAFF1,* and MEINRAT O. ANDREAE2, 1Max Planck Institute for Behavioural Physiology, D-82319 Seewiesen, Germany; 2Max Planck Institute for Chemistry, Biogeochemistry Department, PO Box 3060, D-55020 Mainz, Germany (Manuscript received 9 June 1999; in ﬁnal form 22 December 1999) ABSTRACT Numerous experiments with homing pigeons and other birds strongly suggest that birds dis- placed to unfamiliar remote areas are able to determine their position relative to home by deducing relevant information from atmospheric trace gases perceived by olfaction. These ﬁnd- ings induced the hypothesis that ratios between several airborne compounds show roughly monotonic spatial gradients, diVerently in diVerent directions, over distances of some hundreds of kilometres. To test this hypothesis, 192 air samples were collected, successively in 3 summers, at 96 sites regularly distributed over an area covering a radius of 200 km around Wu ¨ rzburg, Germany. Statistical analysis of the gas chromatographic measurements on these samples revealed that such gradients in the ratios between a number of omnipresent hydrocarbons do in fact exist. The gradients are noisy, but not beyond the range that is compatible with the homing behaviour of pigeons which is noisy as well. The directions of the gradients are remark- ably robust against changes of weather, especially of winds. Winds, however, shift the levels of ratios in the whole area without dramatically changing the directional relationships. A system- atic angular correlation between variations in space and variations caused by winds could theoretically be utilized by birds for navigational purposes. Our analysis dealt mainly with the most abundant anthropogenic hydrocarbons, which are the best-suited tracers to detect spatio- temporal distribution patterns. It is very likely that equivalent patterns exist in naturally emitted volatile compounds as well, given that they are subject to similar variability in the distribution of sources and sinks and similar transport patterns. 1. Introduction a medium they had never expected to be of interest in this context: the atmosphere. Over more than 1.1. T he problem 25 years a large body of evidence has accumulated that strongly suggests that homing pigeons and This investigation was stimulated by navigating other birds are able to deduce spatial information birds. After many years of unsuccessful searches from atmospheric trace gases perceived by olfac- for the physical basis of their ability to home from tion (Papi, 1989, 1991; WallraV, 1990a,b, 1996; unfamiliar distant areas and after exclusion of WallraV et al., 1995; Able, 1996; Bingman, 1998). deductively reasonable sources of positional It is still enigmatic, however, how this might be information, such as the geomagnetic ﬁeld or possible. Doubts about the feasibility of olfactory astronomical clues, researchers were forced by navigation (Schmidt-Koenig and Ganzhorn, 1991; experimental ﬁndings to focus their attention on Wiltschko, 1996; Gould, 1998) ﬁnd immediate belief, because intuitively the atmosphere appears, * Corresponding author. e-mail: [email protected] due to its temporal instability, utterly unsuited as Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1139 a source of spatial information. This preconception which is not individually peculiar but diVers has seemingly been supported by negative reports gradually from the compositions in the neigh- on the suitability of atmospheric trace substances bouring areas. for navigational purposes (Becker and van Raden, (2) Homing of pigeons on an olfactory basis is 1986; Waldvogel, 1987, 1989; Ganzhorn and possible even if the birds lack continuous contact PaVrath, 1995). However, by only considering with environmental air during the outward trip to absolute concentrations of single compounds sep- the release site. Thus, it should be possible to arately, these studies started from restricted pre- determine positions of distant sites relative to the mises which, a priori, could hardly be expected to home site by analyzing ambient air only at these raise positive results. Given the extent of the sites themselves. results on homing, which, as a whole, cannot (3) Pigeons home successfully from sites that plausibly be explained in any alternative way, we may be a hundred or more kilometres away from should trust the pigeons rather than our limited sites they have previously visited. Thus, the knowledge and imaginative competence. Solutions pigeons need not develop a topographical map of can only be expected from a more thorough odours by exploring a large area. Instead, atmo- empirical search for persistent patterns of atmo- spheric conditions appear to allow spatial extra- spheric tracers, particularly with respect to their polations from one site ( home) to an indeﬁnite suitability as a source of positional information. number of other sites in a large surrounding area. In this study, we present the result of the ﬁrst Extrapolations are possible only if there is some investigation of this kind. We show that the atmo- predictable quantitative connection between con- sphere does contain surprisingly persistent spatial ditions at diVerent sites, i.e., if the spatial distribu- structures in the relative proportions of chemical tions of particular airborne trace compounds tracers, structures which potentially provide a include some kinds of fairly monotonic gradients suitable basis for avian navigation. Studies on over longer distances. pigeon homing have thus inspired a search for (4) Since we know that pigeons can ‘smell’ their mesoscale chemical gradients that had not been position before they start ﬂying, they must already previously noticed in the atmosphere. have some knowledge about the directions of gradients without scanning a larger area upon 1.2. Scope of search deduced f rom homing release. Experimental studies suggest that they experiments with pigeons gain this knowledge during their long-term stay at the home site by correlating olfactory conditions Previous experiments with homing pigeons led with the direction of the wind. Normal homing to a working hypothesis predicting that certain behaviour can be observed only in pigeons that properties of the atmospheric chemical environ- were exposed to natural winds at the home site. ment make it suitable as a source of spatial The birds are disoriented and fail to return when information. These hypothetical properties were they previously have been kept in an aviary deduced from the following experimental ﬁndings, screened from winds. Their departure directions conclusions and associated considerations, which can be predictably manipulated by manipulating are all based on the general conclusion that pigeon wind directions in the home aviary. The necessary homing from unfamiliar areas relies on olfactory directional reference is available in the form of a signals (for review of the results themselves, critical sun compass which is known as a part of the valuations, debated interpretations and detailed pigeons’ homing mechanism. references, see Papi (1989, 1991) and WallraV (5) For the pigeons, long-term signal averaging (1990a,b, 1996 )). is possible at one position, the home site. When (1) Pigeons home equally well with both head exposed to the local air at a distant site, they must and tail winds. Thus, they cannot navigate using be able to deduce their position in relation to a chemical signal that is blown directly from the home site to their current position. If, nevertheless, home by comparing actual olfactory conditions pigeon homing is based on odours, the home site with remembered home conditions. The time avail- should be characterized by a certain qualitative able for this comparison is usually limited to, at and quantitative composition of compounds most, a few hours. A period of only 5–10 min, Tellus 52B (2000), 4 1140 h. g. wallraff and m. o. andreae however, seems to be insuYcient to reach full didates also from a physiological point of view, performance. because organisms usually have diYculties in (6) Spatial information deduced from atmo- quantifying absolute intensities. However, animals spheric trace compounds need not be precise. and humans can detect small variations in the Pigeons do usually not ﬂy the shortest way straight proportional composition of a set of odorous home, many fail to home at all, and the routes of substances even with varying levels of absolute a sample of individually ﬂying birds show consid- concentrations. At least in humans, diVerent erable stochastic noise. Orientation performances quantitative proportions between compounds pro- as shown by pigeons can be achieved, on a proba- duce subjectively diVerent qualitative odours. bilistic basis, by evaluating noticeably unreliable The above results and considerations led to the environmental signals. Thus, we need not expect following working hypothesis (WallraV, 1989b, to ﬁnd perfectly monotonic gradients in the atmo- 1991, 1996). A number of trace compounds (min- sphere; considerable noise can be tolerated. imum three), permanently present in the ground- (7) Spatial information deduced from atmo- level air, should systematically vary in their con- spheric trace compounds need not be stable in centrations relative to each other over distances time, as also pigeon homing is temporally highly of several hundreds of kilometres (at least in variable in scales of hours, days, years, and accord- Germany), diVerently in diVerent directions. As a ing to seasons. Performances in winter are much result, any location inside a wide area should be lower than in summer (as are emissions of biogenic characterized by a unique proportional composi- trace gases). tion of concentrations of these compounds (see (8) The spatial range within which pigeons can background stippling in Fig. 1 in WallraV, 1996). deduce their positions relative to home from local This prediction need not be perfectly met; consid- air is fairly large but not unlimited. It seems to erable noise is allowed. To a smaller or larger vary from less than 100 to at least 700 km, appar- degree, the pattern of gradients should systematic- ently depending on geographical ( possibly oro- ally vary with varying directions of the wind. graphical ) conditions. Spatial variations and wind-dependent variations (9) Homing ﬂights of pigeons usually take place should be integratable into a common system. at altitudes of less than 300 m, mostly less than Currently in Central Europe, where many of 100 or 50 m above ground (Hitchcock, 1952; the pigeon homing experiments were conducted, Wagner, 1970). Moreover, pigeons can pick up the most prominent trace gases accessible to gas navigationally useful olfactory information at chromatographic analysis are volatile organic heights of less than 5 m. Air above open land has compounds ( VOCs), mainly hydrocarbons, been demonstrated to contain more reliable released into the atmosphere by human activities, information than air inside forests or between such as vehicular emissions, gasoline production other vegetation. Thus, trace substances to be and handling, and solvent evaporation (Piccot analyzed can be collected close to ground level, et al., 1992; Singh and Zimmermann, 1992; Atlas but air from above open landscapes should be et al., 1993; Fujita et al., 1995; Clarke and Ko, preferred. 1996; Mukund et al., 1996). Thus, the atmosphere (10) Pigeons deduce navigational signals from investigated today diVers considerably from the airborne molecules in the gas phase. Removal of atmosphere in which avian navigation systems aerosol particles by ﬁltration does not interfere evolved. However, this apparent disadvantage may with navigation, whereas scrubbing by charcoal provide a methodological advantage. Due to their eliminates positional information. higher concentrations, many anthropogenic VOCs can be more reliably detected and quantiﬁed than natural compounds. Our hypothesis includes that 1.3. Analytical approach spatiotemporal distributions of natural and A priori, we did not search for spatial gradients anthropogenic compounds follow similar rules in the absolute concentrations of single com- because they are determined by the same atmo- pounds. Their temporal variability is enormous. spheric processes. If this is true, it must be easier Much less variable are proportional relations to decipher these rules in the latter class of com- between compounds. They are the favoured can- pounds. The following analysis deals with the Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1141 proportional relationships among all substances 12 peripheral sites in succession, 3 on each arm of detected by the applied method without regarding a rectangular cross, with 60 km between the sites their chemical nature and origin. on each radius. Four such crosses at angles of about 22.5° to each other, each including 2 sets of alternating sites on each cross, resulted in 96 sites 2. Material and methods (Fig. 1). In 2 successive rounds, each consisting of 8 trips, each site was visited twice in diVerent 2.1. Collection of material years. In addition, two samples were taken at the centre of the ﬁeld, at intervals of 2–3 days, during Samples of atmospheric trace gases were col- each trip (=32 samples). Sampling sites were in lected using small adsorbent tubes (Supelco open or fairly open rural (rarely semi-rural ) coun- Carbotrap 400, containing Carbotrap F, C, B and try, preferentially on hills rather than in valleys Carboxen 569). The sampling device was set up and usually at some distance from larger forests. on the roof of a mini-van with the inlets of the The centre of the sampling grid was identical with sampling tubes about 4.25 m above the ground. the position of a pigeon loft from which many Sampling duration was 2 h and sampling volume homing experiments had been conducted inside 48 l of air. Before and after sampling, the tubes the same area. were stored in liquid nitrogen. Weather conditions varied considerably, some- 96 rural sites, regularly distributed inside a times within a trip, more often between trips. radius of 200 km around a central position close Wind directions during sampling were fairly evenly to Wu ¨ rzburg, Germany (49°46∞29◊N, 10°00∞50◊E), distributed, but east or west winds were somewhat were visited for sampling during 16 trips in 3 more frequent than winds from north or south successive summers (June to September in (Fig. 2). 1995–1997). Each trip lasted 4–5 days and covered 2.2. Gas chromatography In the laboratory, adsorbed compounds were transferred by thermodesorption directly to the capillary column of a gas chromatograph (Hewlett-Packard 5890 Series II ), in which two non-polar fused silica capillary columns (Supelco SPB-1, Cat. No. 2-5304, 15 m, 0.53 mm ID, 1.5 mm Fig. 1. The 96 sampling sites surrounding the centre near Wu ¨ rzburg, Germany. The radius of the outer circle is 200 km. Letters A–D indicate diVerently oriented rectan- gular crosses along which the sampling trips were per- formed. Equally symbolized sites were visited ﬁrst along Fig. 2. Mean directions and speeds of the wind measured one axis and then along the other. Deviations from a during sampling. Crosses show data obtained at the regular distribution were caused by topographical condi- various peripheral sites, circles show those obtained at tions (open rural areas at higher altitudes rather than in the centre. The directions given for 0 km/h (calm) are valleys were preferred ). meaningless. Tellus 52B (2000), 4 1142 h. g. wallraff and m. o. andreae ﬁlm thickness and Cat. No. 2-4037, 60 m, 0.25 mm taining up to about 350 peaks. The compounds ID, 1.0 mm ﬁlm thickness) were mounted in series. were labelled using a ‘‘C index’’ giving their posi- Desorption was made by heating a Carbotrap tion on the time axis in relation to the reten- tube in 3 steps from 280° to 380°C over a period tion times of n-alkanes (example: C6.5= retention of about 20 min. During this time, helium carrier time halfway between n-hexane and n-heptane). gas was driven through tube and columns using Detected peaks ranged from C3.5 to C20. From a head pressure of 49 psi while the oven was about C10 onwards, however, most of the peaks cooled down to −60°C. Thereafter, starting with were quite small and hence detection, identiﬁca- initial settings of +5°C and 16.2 psi, a slow tem- tion and quantiﬁcation less reliable than in the perature/pressure programme was run with a con- ﬁrst part of each chromatogram. stant ﬂow of approx. 1 ml/min over 420 min, which Chromatograms obtained from diVerent sites at ended at 240°C. Peaks were detected and quanti- diVerent times varied considerably with respect to ﬁed by means of a ﬂame ionization detector (FID). numbers and sizes of peaks. Nevertheless, a typical Since ratios between peak areas inside each chro- pattern of peaks was observed very consistently. matogram were the values of interest, no calib- This pattern was particularly clear in the ﬁrst part rations were necessary. Small variations in of each trace between C3.5 and C8. In this range retention times of diVerent chromatograms were of retention times we found all 16 peaks that could semi-automatically equalized, peak areas integ- be clearly identiﬁed in each of the 224 chromato- rated, and compounds indexed according to their grams (Fig. 3). The following analysis deals mainly standardized retention times. with these 16 compounds, because they show most Each sample provided a chromatogram con- clearly the fundamental results of this study. Fig. 3. Parts of 2 representative chromatograms showing the 16 always and everywhere observed peaks. Numbers indicate their C indices (in bold italics with preﬁx ‘‘C’’: 6 ‘‘top compounds’’ according to Fig. 4B). Names of com- pounds identiﬁed by application of the pure substance are given in the bottom graph. Ordinates are in relative but corresponding units; note the diVerent ranges. Inside the boxes are example calculations giving areas (relative units) and resulting ratios for one pair of substances (see text). Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1143 2.3. Statistical algorithms versus A, accordingly, to a ± 180°). Thus, e as a measure of eccentricity indicates a possible spatial The basic unit used for statistical analyses was gradient with an increasing ratio of A versus B in the amount of one particular substance as a the direction a . percentage of the sum of two or more substances N In some calculations, we considered only the within the same chromatogram (see examples C4.7 directions of sampling sites from the centre, not versus C5.3 in the boxes of Fig. 3). Pairwise coup- their distances. In these cases we computed x= ling of the 16 omnipresent peaks resulted in 120 sin b and y= cos b, where b is the direction of the possible combinations. The percentage of either respective site from the centre (clockwise from peak was calculated for each such combination in north). In analogous calculations, b represented each chromatogram. The resulting percent values the direction of wind during sampling. Data were per compound were then standardized in terms of tested for statistical signiﬁcance using Monte diVerences from the mean of a sample of chromato- Carlo methods as described in Section 3. grams divided by the standard deviation (s.d.). Thereby, each compound was represented with equal weight, independently of its original percent- 3. Results age. Standardized diVerences, d, are expressed in a scale of ±s.d. units per sample, where ‘‘sample’’ 3.1. Statistical evidence for large-scale atmospheric may mean either the whole set of chromatograms gradients obtained at the peripheral sites (n= 192), the 12 3.1.1. T he sample/site mixing approach. With a symmetrical sites visited per trip, or any other limited number of sites and measurements, com- selected part of the whole set. plete uniformity cannot be expected even if there In a next step, the diVerences d were associated is no general bias in irregularly scattered standard- with the spatial distribution of their sites of origin. ized diVerences, d, from the overall mean of a ratio In relation to the centre of the ﬁeld and to north, between given compounds. Even on a pure chance x and y values for each site were determined ( km) basis, the value for eccentricity, e, always deviates and multiplied by the d value obtained for this from zero to some degree. Expectations on this site on the respective day. From W xd and W yd basis can be calculated if the d values are linked (resulting from n chromatograms) a mean vector to the 96 sites at random (the two chromatograms was calculated (Mardia, 1972) with gained at the same actual site were always linked length or eccentricity to the same random site). The cross-hatched histo- gram in Fig. 4A shows the average frequency 2 2 n n distribution of e for the 120 pairs of 16 omnipres- ∑ x d + ∑ y d S i i i i A B A B ent substances resulting from 105 independent i=1 i=1 e= (km) (1) random replications. The ﬁlled columns show how the distribution obtained from the actual linkage and direction between data and sites of their origin deviates from this expectation. The actual distribution is ∑ x d i i far beyond the range that could ever be expected a =arc tan i=1 (2) N on the basis of pure chance. The maximum degree ∑ y d of geographical sorting of positive and negative i i i=1 values of d, as shown in the insert of Fig. 4A, was (degrees clockwise from north). not achieved in any one of the 105 random Since sites were symmetrically distributed and permutations. average d is, per deﬁnition, zero, the resulting The circular insert in Fig. 4A (radius 200 km) vector should also be zero if the deviations from shows the relative portions per site of the highest the average proportion between the two sub- ranking pair of compounds, C4.0 compared with stances were uniformly distributed over the sites. C5.3. Filled circles indicate sites with C4.0 above The more e deviates from zero, the more the ratio and C5.3 below its average per trip, empty circles of substance A (with respect to B) deviates from those where the relation is reversed; sizes of circles its overall mean towards the direction a (and B are proportional to absolute standardized diVer- Tellus 52B (2000), 4 1144 h. g. wallraff and m. o. andreae Fig. 4. Geographical eccentricity (e) of means of 16 compounds pairwise compared with each other. A: Frequency distribution of eccentricities of the 120 pairs of substances. Filled columns give the actually observed distribution, cross-hatched columns the average distribution of 105 replications attributing the 192 chromatograms to the 96 sites at random. Thin curves give averages of the 10 (=0.01%) random replications that reached the highest medians (solid line), the best correlations with the original distribution ( broken line) and the highest mean for the uppermost 5 values per run (dotted line). The ﬁlled dot at 16.5 km symbolizes the median of the actually observed distribution, the dot at 7.7 km the overall random median with the 99.9% range of individual medians. The circular insert shows the spatial distribution of ratios of the highest ranking pair of compounds (see text). B: The 16 substances ranked according to the mean eccentricity resulting from pairwise comparison with each of the other 15 substances. Actual values are shown by the uppermost thick curve. The other curves give chance expectations as resulting from 105 replications in which chromatograms were attributed to sites at random. In each run, separate ranking was made according to mean eccentricity (independently of the substance order). The overall random median is shown by the lowermost curve, the upper limits for 95, 99 and 99.9% of the random runs by the three thinner curves. Open unconnected circles indicate lengths of the second-order mean vectors calculated from every 15 e values under consideration of direction a (if all 15 a values were identical, results were identical with the ﬁlled circles). The N N associated second-order mean direction is given along the abscissa together with the C index of the respective substance. Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1145 ences d. Ellipses indicate means and 95% conﬁd- (see horizontal lines 99% and 95% in Fig. 4B) ence areas of the ‘‘black’’ and the ‘‘white’’ sub- and down to rank 9 in less than 8%. Thereafter, sample (one mirrored to the other would result in these chance probabilities increase from 10% to the common eccentricity e= 37.9 km, direction 28% in ranks 10–13 and ﬁnally to 79–83% in a = 309° for C4.0 and 129° for C5.3 ). ranks 14–16. The sequence in these lower ranks The shape of the black histogram in Fig. 4A, is, therefore, uncertain, but identiﬁcation of the reﬂecting the actually observed frequencies, sug- highest-ranking substances should be quite gests a continuous transition from irregularity to reliable. spatial distributions showing a smaller or larger The computational methods we used could be degree of regular directionality. The top values modiﬁed in many details without substantially are not exceptional cases but rather mark the aVecting the principal outcome. Besides the ‘mean upper end of a fairly normal distribution. If vector approach’, as described above, we applied random samples were selected to contain particu- a ‘rotating axis approach’, with similar results. In larly high values, the mass of values remained in this case, an axis running through the centre is the typical random range (Fig. 4A, dotted curve). rotated stepwise through 180°, and in each posi- The real data indicate, in contrast, a general trend tion the cosine of each site projected on the axis towards a larger-scale spatial order in the (i.e., the distance along the axis) acts as x in a atmosphere. regression analysis, while the d values act as y. It is important, but not suYcient, to know that The axis obtaining the highest correlation coeY- some ratios between substances vary with some cient is very similar to the above a , and the directional regularity. Fig. 4B gives, in addition, distribution of maximum correlation coeYcients an individual ranking of the 16 compounds is similar to that of e (although in a diVerent together with the direction of the average vector scale). If the whole sample of 192 chromatograms calculated for each of them. The non-connected is split into two parts (e.g., according to months open symbols in this graph indicate that some of or time of day, according to the two rounds with the compounds (ranks 1, 2, 3, 5, 13) show very visits of the same 96 sites, or sites alternating so consistent ( but diVerent) unidirectional vectors, that no site is included in both sub-samples), the while others (ranks 4, 9, 11) are quite variable in results diVer in details, statistical signiﬁcances tend their directedness depending on the substances (naturally) to be lower, but the overall outcome with which they are compared. These latter com- remains unaVected. pounds constitute a fairly neutral background against which the spatially variable compounds 3.1.2. T he error adding approach. A second, vary in their proportional ratio. completely diVerent approach to testing the reality Such conclusions, however, should not be of gradients is based on the consideration that, in extended, without reservation, to all 16 com- the case of pure random eVects, other proportional pounds we investigated in detail. Although the distributions of peaks within each chromatogram data set as a whole is incompatible with a pure should also produce similar levels of eccentricity. random origin, irregular variability may be inter- To test this possibility, a random error value in mingled with fairly regular trends. We must expect the range of ±s was added to each peak-area that the mean eccentricity levels of the individual value, where s is the standard deviation of a compounds are co-determined by chance. normal distribution in percent of the peak area. Although, for instance, the chance probability is Thus, with increasing s, original peak sizes were less than 0.1% that a compound ranking at posi- increasingly modiﬁed at random to either smaller tion 13 has a mean e value of 14.1 km, this value or larger values. Fig. 5A shows that, along with is reached or exceeded by the compound ranking this increase of s, the average eccentricity of the at position 1 in 28% of the random runs and in uppermost 10% of the 120 pairs of substances some rare cases even by the rank 2 or 3 com- decreased. Beyond of s=±40%, when the evalu- pounds. The probability of chance ranking is quite ated peak areas reﬂect only little of their original low, however, in the highest ranks. The e level of proportions, the original level of eccentricity e was ranks 1–3 has been reached in less than 1% of never reached by any random set of manipulated the random runs, down to rank 5 in less than 5% data. With smaller ranges of superimposed error, Tellus 52B (2000), 4 1146 h. g. wallraff and m. o. andreae Fig. 5. Mean eccentricity of the 12 highest ranking pairs of substances, out of 120 pairs, with random errors superim- posed on the individual peak areas in the 192 chromatograms. A: The heavy curve gives the original (s=0 ) and the mean of 104 random replications with increasing ranges of random errors superimposed on the peak areas. Ranges for 95, 99 and 99.9% of each of the 11 sets of 104 random runs are drawn by thin lines. B: Mean diVerences in average eccentricity between all 16 substances modiﬁed by superimposed random errors (s=±50%; result per run set to zero) and one substance (C index on the abscissa) left in its original state in else identical random runs. 100 replications per selected substance; means and conﬁdence intervals 95, 99 and 99.9%. slightly increased eccentricities were achieved in a repeated with s=±50%, but the peaks of only small percentage of runs, but the average started 15 compounds were modiﬁed by an artiﬁcial to decrease already with about s=±5%, which random error, whereas the peaks of one compound may be the approximate error range of actual were left at their original sizes. In identically measurement. randomized runs, one with 16 and one with only This approach also allows us to rank the indi- 15 compounds manipulated, the diVerence De vidual compounds (Fig. 5B). Calculations were between average e of the uppermost 12 of 120 Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1147 peak pairs was determined. The results of such 3.2. Multi-compound gradient ﬁelds double runs, with each compound successively left Considering not absolute values but ratios, pair- in its original state, revealed a ranking of sub- wise combination of only 2 compounds, resulting stances that is identical in its ﬁrst 6 positions with in one ratio per site, would be insuYcient for those found with the ﬁrst procedure (compare position ﬁxing in a 2-dimensional plane. A min- Fig. 4B). For these 6 substances, the diVerence imum of 3 compounds would be necessary, whose from pure random results in Fig. 5B is so large, ratios entail two degrees of freedom, but with that a general chance basis can be excluded. noisy signals a larger number of gradients would lead to more reliable conclusions (WallraV, 1989a). 3.1.3. A look at not always recorded substances. Combining the 16 substances in 560 diVerent The 6 top substances, selected from a total of triplets yields similarly large diVerences of spatial more than 300 observed peaks by the above regularities from random expectations as the pair- algorithms, demonstrate the existence of diVerently wise combinations shown in Fig. 4. The same is oriented spatial atmospheric gradients, but do not true for higher-order combinations of up to 6 necessarily represent the real top group of sub- compounds. stances with the strongest spatial gradients. In When the 6 highest-ranking substances (Figs. order to avoid any possible manipulation of data, 4B and 5B) are combined, the relative portion of peak integration in the chromatograms followed at least ﬁve of them shows an overall spatial an automated procedure throughout without indi- gradient which is far from simple monotony but vidual optimization. Consequently, quantiﬁcation clearly diVerent in its direction among the sub- and even identiﬁcation of smaller peaks, particu- stances (Fig. 6). Various other combinations (trip- larly when superimposing each other, was often lets, quadruplets, etc.) lead to basically similar associated with considerable ranges of possible ‘chemical landscapes’. It should be noticed that error. Since absolute concentrations of all com- these patterns represent neither snapshots of a pounds varied extremely, the less abundant com- momentary state nor long-term averages, but pounds were not detected in the more dilute rather result from a mixture of many spot checks chromatograms at all, but may, nevertheless, have obtained under quite variable weather conditions been present at very low concentrations. When and wind directions. These conditions were fairly the threshold for inclusion of a substance in the evenly distributed over the whole area. Due to the analysis was set as being observed in at least 80% symmetry of each trip, the observed spatial pat- of the chromatograms, 72 substances met this terns are unlikely to have been caused by hidden criterion. The highest ranking 18 compounds temporal or condition-dependent biases. (25%), in descending order, were then the follow- ing (those already included in the set of 16 are in brackets; e values of the ﬁrst 8 are particularly 3.3. EVects of wind direction and origin of air outstanding): C7.5, C6.9, (C5.1), (C5.3), (C4.0), masses C17.3, C17.6, C14.8 — (C4.7), C5.58, (C6.6), C13.2, 3.3.1. L ittle inﬂuence on directions of spatial C11.5, C18.5, C7.2, C18.0, C19.9, (C3.5). This gradients. When considering potential spatial gra- collection suggests that spatial proportional gradi- dients of trace gases, winds deserve particular ents do not only occur in the lighter compounds attention in a ‘negative’ as well as in a ‘positive’ below C8, as considered in this study in detail, way. The ‘negative’ aspect is quite obvious and but also in substances with a considerably large the main reason why navigation based on airborne molecular weight. As quantities of these substances odours is often presumed unfeasible. In many were always quite small, however, they could not regions, and particularly in those in which homing be recorded and evaluated reliably by the methods of birds has been shown to occur, winds are we applied ( but might be detected and evaluated extremely variable. The atmosphere covering a by birds). It seems very likely that the present given area is rapidly exchanged by air masses investigation, due to its methodological limita- arriving from diVerent directions over diVerent tions, underestimates the frequency and level of types of land or sea. Intuitively and with some spatial regularities in the chemical composition of good will, one might consider it possible that a the atmosphere. certain pattern of gradients is fairly consistently Tellus 52B (2000), 4 1148 h. g. wallraff and m. o. andreae Fig. 6. Proportional relationships between the six compounds ranking highest in Figs 4B and 5B. In each of the 192 chromatograms the peak-area sum of all 6 substances was set to 100 and percentages of the individual compounds determined. For each substance, diVerences from the mean percentage per trip were calculated and transformed into a scale of standard deviation. From the resulting two-dimensional point diagrams (analogous to the insert in Fig. 4A), complete ‘‘landscaspes’’ were computed by interpolation. Areas with values above average are bright, those with values below average are dark. Small diagrams show means± s.e.m. parallel to the optimal axis whose direction is indicated. The abscissa gives the distance from the line running perpendicularly to this axis through the centre; the ordinate gives the diVerence from the overall mean per substance in terms of standard deviation. Distance classes for averaging are 0–60 km, 60–120 km, and 120–200 km. restored each time when winds are blowing, for to the gradients of proportions between our six instance, from the east. With westerly winds, how- top substances selected according to their spatial ever, one would expect a quite diVerent pattern. regularities (Fig. 7). If the sample of 167 chromato- This expectation has not been met with respect grams collected under conditions of wind speed Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1149 Fig. 7. Mean percentages of the 6 top substances ( Fig. 6 ) in various directions from the centre with varying directions of the wind during air sampling (wind speed 5km/h). Directions of sites from the centre (irrespective of distance) were combined in angular classes and running means of the percentage per compound determined in steps of 5° around the circle (unequal angular distribution of sites thereby compensated by averaging). Curves were calculated for two pairs of mutually exclusive sub-sets of chromatograms: with winds from semicircles north and south (thick curves without dots) and semicircles west and east (thin curves with dots). Numbers of included chromatograms are given in brackets. Site-independent means per wind class are shown by thin horizontal lines. Original, non-standard- ized percentages (ordinates) indicate the ranges of variability for each substance. of at least 5 km/h was subdivided into two sub- tion of sites in a roughly similar way in both sub- samples obtained with winds from the eastern or samples (Fig. 7: two dotted curves per diagram). the western semicircle, respectively, the propor- The same was true, if winds from the northern tions varied in dependence on the spatial distribu- and the southern semicircle were compared (two Tellus 52B (2000), 4 1150 h. g. wallraff and m. o. andreae Fig. 8. Mean percentages of the 6 top substances (Fig. 6 ) in various directions from the centre with air masses of diVerent origin as indicated. Graphs are analogous to those in Fig. 7. thick curves). In each substance, the proportional maritime air masses (types mT, mTp, mPt, mP, maximum and minimum was located in similar mPa) and continental air masses (cP, cPt, cTp), spatial directions independently of whether the respectively, are fairly similar to those obtained wind came preferentially from north, south, west with westerly and easterly winds, respectively or east. (Fig. 8). This similarity is largely based on the fact Analogous subdivisions were made with respect that in 77% of the cases winds from the western to the type of air mass as categorized in the semicircle were correlated with maritime air weather reports for each day (Scherhag, 1948), masses and winds from east with continental air with similar results. Curves obtained for days with masses. No such correlation existed between winds Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1151 from north and ‘polar’ air masses (cP, cPt, 192 chromatograms was reduced to 167. On the mPt, mP, mPa) and between winds from south other hand, the 32 chromatograms obtained from and ‘tropical’ air masses (cTp, mT, mTp). collection at the centre ( 22 with wind speed Nevertheless, the curves calculated for these two 5km/h) could be included, so that a total of categories of air masses also roughly ﬁt into the 189 chromatograms could be analyzed with general patterns that are typical for the diVerent respect to a possible eVect of wind direction compounds (Fig. 8). (Fig. 2). As Fig. 9A shows, such an eVect appears The fairly uniform shapes of the curves for each to exist. The distribution of the resulting vector compound in Figs. 7 and 8 indicate that the lengths diVers markedly from random expectation; directions of the gradients were not very much chance probability for the longer vectors to occur modiﬁed by varying wind conditions and air is p< 0.001. However, the top values did not masses. This does not mean, however, that winds wholly reach the level found in analogous calcula- and air masses had no inﬂuence at all. The average tions of spatial vectors (p< 0.0001). portion of each compound in the whole area, but Fig. 9B shows the results for the individual not its directional slope in relation to the other compounds. The data base is less complete than compounds, did vary with wind direction and with in Fig. 9A, but space-dependent and wind-depend- air mass type (see horizontal lines in Figs. 7 and 8). ent values are more immediately comparable, as both calculations comprise the same 167 data sets 3.3.2. Inﬂuence on the level of ratios between obtained at the peripheral sites with wind speeds compounds. This latter facet leads to the potential 5km/h. Both aspects interfere and ‘‘compete’’ ‘positive’ aspect of winds, i.e., to their assumed with each other: EVects of winds are intermingled functional involvement in bird navigation. In with spatial eVects and vice versa. Fig. 9B suggests order to derive, at an unfamiliar remote site, that spatial regularities are more clearly expressed, information on the current position relative to and statistically better signiﬁcant, than wind-con- home from proportions between compounds, birds nected regularities. The ordering of substances need to know the direction of the gradient of each listed along the abscissa in Fig. 9B shows that of the compounds of which they make use. As there is, by and large, a negative correlation mentioned above, homing experiments suggest the between space dependence and wind dependence hypothesis that pigeons are able to deduce these of the individual substances. Five of the six com- directions, without leaving the home site, from pounds composing the top group under spatial changes of compound ratios that are correlated aspects (Figs. 4B, 5B) are assembled at the lower with changes of wind direction. Thus, it is neces- end when ranked according to their regular sary to investigate whether such correlations do dependence on wind direction. Thus, the sub- exist, and if they do, which relationships they stances showing the clearest spatial gradients in show to the spatial gradients of proportions. their proportional presence appear less aVected Wind directions were used in our calculations by varying winds. in essentially the same way as described for the spatial directions. Instead of the compass direction 3.4. Relationships between spatial gradients and of each site from the centre, the direction of wind wind-related variations measured during air sampling was used to com- pute mean vectors. Distance from the centre could Finally, we look for directional relationships have been replaced by wind speed, but was not, between spatial gradients and shifts of ratios because possible correlations were expected much between substances depending on wind direction. more with directions rather than with the more For this purpose, we mimic the situation of pigeons variable speeds of winds during the short period whose loft is in the centre of our sampling ﬁeld. of measurement (and in fact, results were less clear During their long stay there, and before any when speeds were additionally included ). displacement, birds are assumed to associate the However, a threshold of minimum speed was set proportional composition of a set of substances at 5 km/h, because with lower speeds also direc- with current wind directions. Unlike the pigeons, tions were unreliable or could not be determined we do not have continuous long-term measure- at all. With this threshold, the original sample of ments ‘at home’ but have only a limited number Tellus 52B (2000), 4 1152 h. g. wallraff and m. o. andreae Fig. 9. Wind-dependent eccentricity of means of 16 compounds pairwise compared with each other. Graphs ana- logous to Fig. 4. Histograms and solid lines refer to wind data, broken lines (for comparison) to analogously computed spatial data. Only directions were considered, not speeds of wind and distances of sites from the centre (scale for vector length is the unit circle with a maximum of 1 ). Random expectations were calculated from 104 replications. Thin lines in A give means of the 10 (=0.1%) best random runs. Numbers along the abscissa in B are C indices in rank-order sequence for winds (above) and sites ( below); the six compounds ranking highest in Figs. 4B and 5B are in italics. Corresponding mean directions (degrees clockwise from north) for winds (above) and sites ( below) along the top of B. For further explanations see Fig. 4; for selection of data see text. of spot checks. Nevertheless, our data imply a pairs of substances, for triplets or for larger groups, general systematic relationship between the direc- for the 16 omnipresent compounds or for larger tions of spatial gradients and the directions of sets. The result is, in principle, always the same. maximum changes of ratios depending on winds. Fig. 10A shows the angular relationships derived Both kinds of directions can be calculated for from the 2556 pairwise combinations that are Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1153 Fig. 10. Angular relationships between corresponding directions of wind-dependent and site-dependent eccentricities of compound ratios. A: Results obtained from pairwise combinations of 72 substances. Cross-hatched columns and curves show deviations of all directions of average increase depending on wind from the corresponding spatial gradient (n= 2556 pairs of compounds). Filled columns include only the more distinct eccentricities (e> 0.075 for both space and wind; n=525 pairs). Histograms refer to all wind data (5km/h; n=189 samples), curves only to winds at the peripheral sites (solid line; n= 167 samples) or at the central site ( broken line; n=22 samples), respect- ively. Spatial data include always the total of 192 samples collected at the peripheral sites. B: Deviations of ‘wind slope’ from ‘spatial slope’ of the 16 individual omnipresent compounds, each paired with its 15 partners. Triangles give mean vectors, each calculated from 15 directions (ordinate 0= uniformly distributed, 1=all identical ). The 6 top substances ( Fig. 6 ) are marked by their C indices and by ﬁlled triangles. possible with the 72 compounds detected in at (data sample for centre quite small ) or eVects of least 80% of the air samples. In most cases, the local conditions at the central site. steepest increase in the ratio of a substance in For most compounds, the wind mean deviates relation to wind direction occurs in a direction clockwise from the spatial mean, but the degree deviating clockwise from its steepest spatial gradi- of divergence is quite variable (Fig. 10B). C5.3, for ent. The mean deviation is about 50°. The fairly instance, one of the ‘‘top substances’’, shows an good agreement between wind-related directions angle diVering only slightly from zero. The propor- calculated from samples taken at the central site tion of this substance tends to decrease geograph- and at the peripheral sites (curves in Fig. 10A) ically from about SE to NW (Fig. 4), and makes it highly unlikely that the result is due to correspondingly, it tends to be highest with winds chance. The slight diVerence between the two from SE and lowest with winds from NW. In curves may indicate either minor eVects of chance contrast, C4.0, another ‘‘top substance’’, spatially Tellus 52B (2000), 4 1154 h. g. wallraff and m. o. andreae increasing from SE towards NW (Fig. 4), showed ized regions of the globe. The most important its highest ratio not with winds from NW, but sources are transport related; tailpipe emissions, with winds from NE. gasoline evaporation and production account for These results suggest that there is a systematic over half the VOCs found in urban and rural directional relationship between relative amounts regions in North America and Europe (Lin and of substances as correlated with geographical posi- Milford, 1994; Clarke and Ko, 1996; ScheV et al., tion and with wind direction. According to the 1996). In Germany, anthropogenic emissions of available data, this relationship is not so simple VOCs (not including methane) are thought to that, for instance, with northerly winds a relative exceed biogenic ones by a factor of 2–10 (Guenther increase of those substances occurs whose relative et al., 1995; Simpson et al., 1995). The substances portion is spatially larger in the north than in the that in our study showed strong spatial or wind- south. Instead, northerly winds seem to cause a related structure can be related to a variety of relative increase of substances that are spatially emission sources: The butanes (C3.5, C4.0) are most abundant in the northwest or even in the emitted during natural gas handling, gasoline west. The generally observed clockwise shift of evaporation and reﬁning. The pentanes (C4.7), the ‘‘wind’’ against ‘‘space’’ might be a consequence of hexanes (C5.3, C5.69, C5.8), benzene (C6.6) and the fact that air masses usually follow curved toluene (C7.7) are characteristic of automobile trajectories. In an anticyclonic system, for instance, exhaust. Toluene is emitted also during the use of northerly winds are connected with air masses a variety of coatings (paints). Methyl chloroform arriving from northwesterly areas and hence might (C6.5) is widely used as a degreasing solvent, increase the relative portion of ‘‘northwesterly’’ although its use is now declining since it was substances. Retrospective counts revealed that our banned under the Montreal Protocol. data samples were not collected under equilibrated Due to the uneven geographic distribution of synoptic weather conditions according to Hess source types, the emission ﬁelds of the diVerent and Brezowsky (1977). As far as the days could VOCs in Europe show distinct gradients at scales be adequately categorized under this aspect, 80% of hundreds to thousands of kilometres (B. of the 164 samples we obtained at the peripheral Langmann, pers. comm., 1998). Consequently, air sites were taken during anticyclonic conditions masses arriving from the west and northwest and only 20% in cyclonic conditions. Among the would have passed over long fetches with high 19 usable central-point measurements 84% were emission densities for many of the vehicle-related anticyclonic and 16% cyclonic. In contrast, the VOCs, whereas air masses from easterly directions percentages for the full periods June–September would have been exposed more to emissions from 1995–1997 diVered much less: 58% anticyclonic coal burning. The emissions of biogenic VOCs, and 42% cyclonic. So we cannot decide whether such as isoprene and the monoterpenes, also show the directional relationships shown in Fig. 10 are pronounced geographic gradients. Isoprene emis- generally representative or valid only for certain sions increase by almost an order of magnitude weather conditions. The few data available for from the centre of the study area to a region some cyclonic conditions result in less clearly expressed, 300 km to the WSW, where Europe’s highest but even larger, counterclockwise deviations. So isoprene emissions are occurring (Simpson et al., far, therefore, it cannot be excluded that long- 1995). Because of technical limitations, isoprene term continuous observation might produce an was the only biogenic substance for which we average angular divergence between ‘‘wind’’ and were able to observe signiﬁcant spatial gradients. ‘‘space’’ not very far from zero. However, given the vast diversity of biogenic volatile substances emitted by diVerent species of plants, animals, and microbes (Isidorov et al., 3.5. Chemical identity of the atmospheric tracers 1985), and the close linkage of these emissions to The most prominent substances in our chroma- taxonomic and ecological parameters (Monson tograms are, with the exception of isoprene (C5.1), et al., 1995), it is highly plausible that similar all of predominantly anthropogenic origin. This large-scale gradients exist in other biogenic VOCs situation reﬂects the overwhelming inﬂuence of as well. It has been shown, for example, that the anthropogenic VOC emissions in the industrial- highest monoterpene emissions are in northern Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1155 Europe, where coniferous trees are most abundant accomplished a ﬁrst step from pure speculation to (Guenther et al., 1995). empirical research. Based on the empirical data The inhomogeneous spatial distribution of VOC described above, a computer model (WallraV, sources provides one of the reasons why these 1989b) can be applied by which navigation from substances may show spatial gradients in their any peripheral point to the centre is possible relative concentrations. But even in the absence without previous knowledge of proportional com- of diVerential source patterns, gradients in relative positions in the distant areas, provided that the abundance would develop due to the diVerent rough directions of the gradients are known atmospheric lifetimes of the individual species. The (WallraV, 2000 ). The results shown are consistent lifetimes of the alkanes and aromatics, which are with the hypothesis that this necessary knowledge most prominent in our data set, typically fall in can be obtained, at the birds’ home site, from a the range of 1–7 days, while the lifetimes of correlation between wind direction and propor- isoprene and the monoterpenes are of the order tional composition of a set of airborne odorants. of hours to a couple of days. However, even when It is not yet clear whether this correlation can be these substances are oxidized, they persist in the as simple as hitherto presupposed by the model atmosphere in the form of oxidation products, or whether a more complicated mechanism must which may have considerably longer lifetimes. be assumed. Many more data, simultaneously Considering that lifetimes of a day or two at wind collected over an extended grid of sampling sites, speeds around 10–20 km/h correspond to would be required in order to fully understand enfolding distances of about 200 to 1000 km, it is the interrelations between spatial ratio gradients quite plausible that a combination of diVerential and changes of ratios in dependence on wind source distribution and chemical conversion in directions. In this respect, our data set must be the atmosphere creates spatial gradients on the considered preliminary. scales relevant for navigation of birds. The exist- Our study did not attempt to identify airborne ence of analogous source inhomogeneity and substances that pigeons and other birds actually diVerential reactivity for anthropogenic and nat- use for navigation. However, we have determined ural substances, as discussed above, together with that there are compounds in the atmosphere which the fact that we found gradient structures also for theoretically could be used. One of these sub- isoprene, suggest that it is reasonable to generalize stances was identiﬁed as isoprene, a hydrocarbon our ﬁndings to natural VOCs. emitted by plants. The fact that most of the other substances investigated here cannot be identical, due to their anthropogenic origin, with those to 4. Discussion which avian navigation systems have been adapted during evolution, does not preclude the inference The principal aim of this study was to explore that atmospheric conditions make olfactory nav- whether the atmosphere contains trace gases igation basically feasible. Gradual variations in whose proportional distribution yields fairly con- the ratios of compounds are likely to result from sistent gradients covering several hundreds of kilo- a combination of structured source patterns, vary- metres in various directions. Statistical analysis ing speeds of chemical conversion and turbulent has revealed that this question can be answered horizontal and vertical mixing. In spite of advect- in the aYrmative (Figs. 4–6). Directions of spatial ive transport with the prevailing winds, such gradi- gradients are noisy but remarkably robust against ents may persist over large regions because of the changes of weather, especially of winds (Fig. 7) turbulent character of horizontal and vertical and origin of air masses (Fig. 8). Nevertheless, diVusion in the atmosphere. By advective trans- ratios between substances show a dependence on port with the prevailing winds, such gradients may wind direction as well (Fig. 9). A systematic rela- be displaced as a whole while their directions tionship exists between variations in space and roughly persist over large regions (cf. Figs. 7 and variations correlated with the direction of winds 8). In contrast to the nonspeciﬁc method applied (Fig. 10). here, which could not be focused on particular With these ﬁndings, the search for the atmo- substances known in advance, birds are probably spheric basis of avian olfactory navigation has specialized and most sensitive to a few most Tellus 52B (2000), 4 1156 h. g. wallraff and m. o. andreae suitable compounds which, for methodological atmospheric structures exploitable for navigation reasons, could not be included in the above ana- do exist. Science in general, however, should not lysis but might provide more reliable spatial gradi- be satisﬁed. It should feel challenged to ask how ents than those revealed by the crude means the observed spatio-temporal regularities in the proportional composition of airborne trace sub- employed in this ﬁrst approach. stances originate and how they achieve their relat- Regional climatic and geomorphological fac- ive stability. In particular, two avenues should be tors, and factors dependent on them, probably explored: First, the simulation of the atmospheric determine the sizes of suYciently monotonic ﬁelds distributions of speciﬁc VOCs in regional models, of ratio gradients. It is to be expected that on the using known source distribution ﬁelds, should be European continent, with its small-scaled coastal used to obtain gridded data sets which can be and mountainous structures and rapidly changing subjected to the same statistical analysis as applied weather conditions, sizes of gradient ﬁelds are to the ﬁeld data used in our study. This would limited and regionally variable. If, nevertheless, make it possible to investigate which physical and remarkably stable and relatively extended gradient chemical processes are involved in creating the ﬁelds exist even there, one may expect much more structures we observed. Second, observational extended roughly monotonic gradients over and studies involving synchronous sampling over a inside the oceans, where air and water currents grid of sites should be designed based on the are much more regular. Conditions for large-range results from these modeling exercises. These olfactory navigation might be particularly favour- experiments may provide a critical test of how able for marine animals such as albatrosses, pet- well the interactions of advective and turbulent rels, turtles, whales, salmons and other ﬁshes, transport, as well as source and sink processes, which are known to migrate and navigate over are represented in regional atmospheric models. distances of thousands of kilometres over or in the open sea (Quinn and Dittman, 1992; Papi and Luschi, 1996; Lohmann et al., 1999). Olfactory 5. Acknowledgements landscapes over seemingly featureless oceans, of which sea birds apparently make use in smell- We thank G. Schebeske for his help in estab- guided foraging ﬂights (Nevitt et al., 1995; Nevitt, lishing gas chromatographic techniques. We are 1999), could be much more eVectively utilized for grateful to B. Langmann for providing us with goal ﬁnding over long distances, if they would information on the spatial distribution of hydro- include extended ratio gradients. carbon emissions in Europe. Data on air masses Birds as well as biologists investigating bird and synoptic weather conditions were kindly pro- homing can be satisﬁed by the apparent fact that vided by Deutscher Wetterdienst, Munich. REFERENCES Able, K. P. 1996. The debate over olfactory navigation Fujita, E. M., Watson, J. G., Chow, J. C. and Magliano, by homing pigeons. J. Exp. Biol. 199, 121–124. K. L. 1995. Receptor model and emissions inventory Atlas, E. L., Li, S.-M., Standley, L. J. and Hites, R. A. source apportionments of nonmethane organic gases 1993. Natural and anthropogenic organic compounds in California San Joaquin Valley and San Francisco in the global atmosphere. In: Global atmospheric chem- Bay area. Atmos. Environ. 29, 3019–3035. ical change (eds. C. N. Hewitt and W. T. Sturges). Ganzhorn, J. U. and PaVrath, D. 1995. Patterns in air Elsevier Applied Science, London, pp. 313–381. pollution as model for the physical basis for olfactory Becker, J. and Van Raden, H. 1986. Meteorologische navigation in pigeon homing. J. Ornithol. 136, Gesichtspunkte zur olfaktorischen Navigationshypo- 159–165. these. J. Ornithol. 127, 1–8. Gould, J. L. 1998. Sensory bases of navigation. Current Bingman, V. P. 1998. Spatial representations and homing Biology 8, R731–R738. pigeon navigation. In: Spatial representation in animals Guenther, A., Hewitt, C. N., Erickson, D., Fall, R., Geron, (ed. S. Healy). Oxford Univ. Press, Oxford, pp. 69–85. C., Graedel, T., Harley, P., Klinger, L., Lerdau, M., Clarke, A. G. and Ko, Y. H. 1996. The relative signiﬁc- Mckay, W. A., Pierce, T., Scholes, B., Steinbrecher, R., ance of vehicular emissions and other emissions of Tallamraju, R., Taylor, J. and Zimmerman, P. 1995. A volatile organic compounds in the urban area of Leeds, global model of natural volatile organic compound UK. Science of the total environment 190, 401–407. emissions. J. Geophys. Res. 100, 8873–8892. Tellus 52B (2000), 4 spatial gradients in atmospheric trace gases 1157 Hess, P. and Brezowsky, H. 1977. Katalog der Großwet- Scherhag, R. 1948. Neue Methoden der Wetteranalyse und terlagen Europas. Ber. Dtsch. Wetterdienst. 15, no. Wetterprognose. Springer, Berlin. Schmidt-Koenig, K. and Ganzhorn, J. U. 1991. On the 113, 1–14. problem of bird navigation. Persp. Ethol. 9, 261–283. Hitchcock, H. B. 1952. Airplane observations of homing Simpson, D., Guenther, A., Hewitt, C. N. and Stein- pigeons. Proc. Amer. Phil. Soc. 96, 272–289. brecher, R. 1995. Biogenic emissions in Europe (1 ). Isidorov, V. A., Zenkevich, I. G. and IoVe, B. V. 1985. Estimates and uncertainties. J. Geophys. Res. 100, Volatile organic compounds in the atmosphere of for- 22,875–22,890. ests. Atmos. Environ. 19, 1–8. Singh, H. B. and Zimmerman, P. B. 1992. Atmospheric Lin, C. and Milford, J. B. 1994. Decay-adjusted chemical distribution and sources of nonmethane hydrocar- mass balance receptor modeling for volatile organic bons. In: Gaseous pollutants: characterization and cyc- compounds. Atmos. Environ. 28, 3261–3276. ling (ed. J. O. Nriagu). John Wiley, New York, Lohmann, K. J., Hester, J. T. and Lohmann, C. M.F. pp. 177–235. 1999. Long-distance navigation in sea turtles. Ethol. Wagner, G. 1970. Verfolgung von Brieftauben im Heli- Ecol. Evol. 11, 1–23. kopter. Rev. Suisse Zool. 77, 39–60. Mardia, K. V. 1972. Statistics of directional data. Aca- Waldvogel, J. A. 1987. Olfactory navigation in homing demic Press, London. pigeons: are the current models atmospherically real- Monson, R. K., Lerdau, M. T., Sharkey, T. D., Schimel, istic? Auk 104, 369–379. D. S. and Fall, R. 1995. Biological aspects of con- Waldvogel, J. A. 1989. Olfactory orientation in birds. structing volatile organic compound emission invent- Current Ornithology 6, 269–321. ories. Atmos. Environ. 29, 2989–3002. WallraV, H. G. 1989a. Simulated navigation based on Mukund, R., Kelly, T. J. and Spicer, C. W. 1996. Source unreliable sources of information (models on pigeon attribution of ambient air toxic and other VOCs in homing, part 1 ). J. T heor. Biol. 137, 1–19. Columbus, Ohio. Atmos. Environ. 30, 3457–3470. WallraV, H. G. 1989b. Simulated navigation based on Nevitt, G. 1999. Foraging by seabirds on an olfactory assumed gradients of atmospheric trace gases (models on pigeon homing, part 2 ). J. T heor. Biol. 138, landscape. Am. Scientist 87, 46–53. Nevitt, G., Veit, R. R. and Kareiva, P. 1995. Dimethyl 511–528. WallraV, H. G. 1990a. Navigation by homing pigeons. sulﬁde as a foraging cue for Antarctic procellariiform Ethol. Ecol. Evol. 2, 81–115. seabirds. Nature 376, 680–682. WallraV, H. G. 1990b. Long-distance navigation of Papi, F. 1989. Pigeons use olfactory cues to navigate. homing pigeons based on airborne olfactory signals. Ethol. Ecol. Evol. 1, 219–231. In: ISOT X: Proc. 10th Int. Symp. Olfaction and taste Papi, F. 1991. Olfactory navigation. In: Orientation in (ed. K. B. Døving). GCS A/S,Oslo, pp. 26–35. birds (ed. P. Berthold). Birkhauser, Basel, pp. 52–85. WallraV, H. G. 1991. Conceptual approaches to avian Papi, F. and Luschi, P. 1996. Pinpointing ‘‘Isla Meta’’: navigation systems. In: Orientation in birds (ed. P. Ber- the case of sea turtles and albatrosses. J. Exp. Biol. thold). Birkha ¨user, Basel, pp. 128–165. 199, 65–71. WallraV, H. G. 1996. Seven theses on pigeon homing Piccot, S. D., Watson, J. J. and Jones, J. W. 1992. A deduced from empirical ﬁndings. J. Exp. Biol. 199, global inventory of volatile organic compound emis- 105–111. sions from anthropogenic sources. J. Geophys. Res. 97, WallraV, H. G. 2000. Simulated navigation based on 9897–9912. observed gradients of atmospheric trace gases (models Quinn, T. P. and Dittman, A. H. 1992. Fishes. In: Animal on pigeon homing, part 3 ). J. T heor. Biol., in press. homing (ed. F. Papi). Chapman & Hall, London, WallraV, H. G., Kiepenheuer, J., Neumann, M. F. and pp. 145–211. Streng, A. 1995. Homing experiments with starlings ScheV, P. A., Wadden, R. A., Kenski, D. M., Chung, J. deprived of the sense of smell. Condor 97, 20–26. and WolV, G. 1996. Receptor model evaluation of the Wiltschko, R. 1996. The function of olfactory input in southeast Michigan ozone study ambient NMOC pigeon orientation: does it provide navigational measurements. J. Air & Waste Manag. Ass. 46, information or play another role? J. Exp. Biol. 199, 1048–1057. 113–119. Tellus 52B (2000), 4

Journal

Tellus B: Chemical and Physical Meteorology – Taylor & Francis

Published: Jan 1, 2000

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Spatial gradients in ratios of atmospheric trace gases: a study stimulated by experiments on bird navigation

Spatial gradients in ratios of atmospheric trace gases: a study stimulated by experiments on bird navigation

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Spatial gradients in ratios of atmospheric trace gases: a study stimulated by experiments on bird navigation

Spatial gradients in ratios of atmospheric trace gases: a study stimulated by experiments on bird navigation

References (29)

Abstract

Journal

Recommended Articles

There are no references for this article.

Our policy towards the use of cookies