ORIGINAL ARTICLE
Annual pollen traps reveal the complexity of climatic controlon pollen productivity in Europe and the Caucasus
W. O. van der Knaap
 
Jacqueline F. N. van Leeuwen
 
Helena Svitavska´-Svobodova´
 
Irena A. Pidek
 
Eliso Kvavadze
 
Maia Chichinadze
 
Thomas Giesecke
 
Bogusław Michał Kaszewski
 
Florencia Oberli
 
Laimdota Kalnin¸ a
 
Heather S. Pardoe
 
Willy Tinner
 
Brigitta Ammann
Received: 19 November 2009/Accepted: 21 March 2010
 The Author(s) 2010. This article is published with open access at Springerlink.com
Abstract
 Annual PAR (pollen accumulation rates; grainscm
-
2
year
-
1
) were studied with modified Tauber trapssituated in ten regions, in Poland (Roztocze), the CzechRepublic (two regions in Krkonosˇe, two in Sˇumava),Switzerland (4 regions in the Alps), and Georgia (Lag-odekhi). The time-series are 10–16 years long, all endingin 2007. We calculated correlations between pollen dataand climate. Pollen data are PAR summarized per region(4–7 traps selected per region) for each pollen type (9–14per region) using log-transformed, detrended medians.Climate data are monthly temperature and precipitationmeasured at nearby stations, and their averages overall possible 2- to 6-month windows falling within the20-month window ending with August, just prior to theyearly pollen-trap collection. Most PAR/climate relation-ships were found to differ both among pollen types andamong regions, the latter probably due to differencesamong the study regions in the habitats of plant popula-tions. Results shared by a number of regions can be sum-marized as follows. Summer warmth was found to enhance
Communicated by F. Bittmann.W. O. van der Knaap (
&
)
 
 J. F. N. van Leeuwen
 
 F. Oberli
 
W. Tinner
 
 B. AmmannInstitute of Plant Sciences and Oeschger Centre For ClimateChange Research, University of Bern, Altenbergrain 21,3013 Bern, Switzerlande-mail: Pim.vanderKnaap@ips.unibe.chJ. F. N. van Leeuwene-mail: Jacqueline.vanLeeuwen@ips.unibe.chF. Oberlie-mail: pfoberli@bluewin.chW. Tinnere-mail: Willy.Tinner@ips.unibe.chB. Ammanne-mail: Brigitta.Ammann@ips.unibe.chH. Svitavska´-Svobodova´Institute of Botany, Academy of Sciences of the Czech Republic,Castle 252 43, Pru˚honice, Czech Republice-mail: svitavska@ibot.cas.czI. A. Pidek 
 
 B. M. KaszewskiInstitute of Earth Sciences, Maria Curie-Skłodowska Universityin Lublin, al Kras´nicka 2 c/d, 20-718 Lublin, Polande-mail: i.pidek@poczta.umcs.lublin.plB. M. Kaszewskie-mail: boguslaw.kaszewski@poczta.umcs.lublin.plE. Kvavadze
 
 M. ChichinadzeInstitute of Paleobiology, National Museum of Georgia,Niagvris 4, 0108 Tbilisi, Georgiae-mail: ekvavadze@mail.ruM. Chichinadzee-mail: maizdr@yahoo.comT. GieseckeAlbrecht-von-Haller-Institute for Plant Sciences, Departmentof Palynology and Climate Dynamics, University of Go¨ttingen,Untere Karspu¨le 2, 37073 Go¨ttingen, Germanye-mail: Thomas.Giesecke@biologie.uni-goettingen.deL. Kalnin¸ aFaculty of Geography and Earth Sciences, University of Latvia,Rainis bvld. 19, Riga 1586, Latviae-mail: laimdota.kalnina@lu.lvH. S. PardoeDepartment of Biodiversity and Systematic Biology, NationalMuseum Wales, Cathays Park, Cardiff CF10 3NP, UK e-mail: Heather.Pardoe@museumwales.ac.uk 
 1 3
Veget Hist Archaeobot (2010) 19:285–307DOI 10.1007/s00334-010-0250-6
 
the following year’s PAR of 
 Picea
,
 Pinus
 non-
cembra
,
 Larix
 and
 Fagus
. Cool summers, in contrast, increase thePAR of 
 Abies
,
 Alnus viridis
 and Gramineae in the fol-lowing year, while wet summers promote PAR of 
 Quercus
and Gramineae. Wetness and warmth in general werefound to enhance PAR of 
 Salix
. Precipitation was found tobe more important for PAR of 
 Alnus glutinosa
-type thantemperature. Weather did not have an impact on the PARof Gramineae, and possibly of Cyperaceae in the sameyear. Care is advised when extrapolating our results to PARin pollen sequences, because there are large errors associ-ated with PAR from sediments, due to the effects of taphonomy and sedimentation and high uncertainty indating. In addition, in pollen sequences that have decadal tocentennial rather than near-annual resolution, plant-inter-action effects may easily out-weigh the weather signal.
Keywords
 Pollen monitoring
 
Annual pollen accumulation
 
 Influx
 
 Climate
 
Europe
 
 Caucasus
Introduction
The study of pollen deposition with annual pollen trapsduring long series of years is especially valuable for ourunderstanding of the relationships between the pollen sig-nal on one hand, and plant geography, vegetation, envi-ronment, climate and weather on the other (Kvavadze2001; Tinsley 2001; Tonkov et al. 2001; van der Knaap et al. 2001a; Autio and Hicks 2004; Hicks 1974, 1977, 1985; Gerasimidis et al. 2006; Seppa ¨ and Hicks 2006;Barnekow et al. 2007; Jensen et al. 2007; Sjo ¨gren et al.2008). Long time-series of annual pollen trapping are nowdeveloping in many European and a few Asian countries,within the scope of the Pollen Monitoring Programme PMP(INQUA working group; http://pmp.oulu.fi/guide.html;Hicks et al. 1996, 1999) first initiated by Sheila Hicks (Oulu, Finland) and under her guidance. This has resultedin numerous publications based on pollen traps in Finlandwhere pollen-trap work started decades ago (Hicks et al.2004; Ra¨sa¨nen et al. 2004; Hicks and Sunnari 2005; Hicks 2001, 2006). Now that the time-series in many other countries are starting to exceed 10 years we have decidedto report on them. Hicks et al. (2001) presented a firstoverview of spatial variation in arboreal pollen depositionacross European PMP sites. Long data series obtainedwithin the framework of PMP could be successfully cor-related with annual pollen sums obtained by aerobiologicalmonitoring with the use of Burkard traps and other types of volumetric samplers (Levetin et al. 2000; Oikonen et al.2005; Pidek et al. 2006; Ranta and Satri 2007; Ranta et al. 2007, 2008a). This paper deals with the relationships between absolutepollen values and climate variables in four countries in ornear Europe. Related papers deal with the relationshipbetween results of pollen traps and surface pollen collectedclose to the traps (Pardoe et al. 2010), with inter-annualvariation and pollen/vegetation relationships of 
 Fagus
 PAR(Pidek et al. 2010), with absolute pollen productivities of Tertiary relict taxa (Filipova-Marinova et al. 2010), andwith composite pollen-dispersal functions (Sjo¨gren et al.2010). They also deal with the longest pollen-trap time-series in Europe (Nielsen et al. 2010), tree-line detectionwith pollen traps and macrofossil traps (Birks and Bjune2010), while Giesecke et al. (2010) provide an historical overview of the use of pollen traps.The research questions addressed have interfaces withseveral neighbouring fields. Examples are phenology, inwhich the timing of flowering and of leaf bud-burst isstudied in relation to meteorological factors in winter andspring (Menzel 2003; Studer et al. 2005), aerobiology, in which the timing and quantities of allergenic pollen arestudied (pollen calendar) and their relationships to meteo-rological factors of the current year (e.g. Rodriguez de laCruz et al. 2008) and ecophysiology, in which resourceallocation is estimated under varying weather conditions inspecies with and without mast-fruiting (e.g. Ranta et al.2005). This study also contributes to a better understandingof the relationships between pollen productivity andmeteorological parameters (e.g. van der Knaap and vanLeeuwen 2003; Kamenik et al. 2009), which is relevant for attempts at reconstructing summer temperature and annualprecipitation based on pollen on broad scales in space andtime (e.g. Seppa¨ and Birks 2001, 2002; Heikkila ¨ and Seppa¨2003; Giesecke et al. 2008). The aim of this paper is to explore the relationshipsbetween climate and annual pollen-trap PAR (influx) for alarge selection of pollen types in countries other thanFinland with long time-series (10 or more years). PAR(pollen accumulation rates; grains cm
-
2
year
-
1
) haveadvantages over pollen percentages, for instance that PARdoes not suffer from percentage distortions (Moore et al.1991). Pollen traps are exceptional in their capacity todetermine PAR in a reliable way, which is hardly ever thecase in peat or sediment sections because chronologicalcontrol is rarely perfect (e.g., Bennett and Hicks 2005;Goslar et al. 2009) and taphonomic processes are ofteninsufficiently understood.Relatively little is known about the relationships betweenannual weather conditions and pollen productivity of plantpopulations in different field situations. The plant popula-tions that produce the pollen collected in our traps grow in awide range of micro-climatic conditions depending on fac-tors such as elevation and exposure. The habitats of thepollen-source populations may be positioned anywhere in
286 Veget Hist Archaeobot (2010) 19:285307
 1 3
 
the climatic space in which the plants can occur. Further-more, there is the impact of climatic conditions on thetransport of pollen from the source plants to the pollen traps.Due to this complexity, this study is explorative rather thanhypothesis-testing. The aim of our interpretation is to pro-videideasforhypothesesthatmightbetestedatalaterstage,with the use of additional types of field data. We willdevelop a method for dealing with ‘noisy’ pollen-trap data,attempttentativeexplanationsfortherelationshipsrevealed,and explore ecological and geographical trends. We willdiscusswhichdataandwhichinsightsarenew,howandwhyour findings agree or disagree with earlier findings, to whatextent our results are useful for palaeo-ecological studies,and which new scientific questions arise. Finally we willpresent a well founded recommendation as to why this typeof work should be continued.
Methods
Pollen-trap data from Poland, the Czech Republic, Swit-zerland and Georgia are used in this paper to establishpollen/climate relationships. The method is essentially acorrelation between the 10–16 year time-series of annualpollen accumulation rates (PAR) in modified Tauber traps(henceforward pollen traps, or traps) with monthly tem-peratures and precipitation of both the year of pollencollection and the previous year. The pollen data aresummarized according to ten climatically homogeneousregions in the four countries, prior to statistical correlation.The unit of pollen data is basically PAR (pollen grainscm
-
2
year
-
1
; also referred to as pollen influx); percentagesare not used in this study.Pollen-data collectionAnnual pollen accumulation rates were monitored withpollen traps following the PMP guidelines (Hicks et al.1996, 1999). Table 1 lists details of the pollen trap regions in the different countries. The length of the time-seriesvaries from 10 to 16 years, all ending with the year 2007.Initially, data from six countries were considered; twocountries (Latvia; U.K.) have been dropped as the pollendata were inconsistent or incomplete (Fig. 1). A selectionof pollen types was used, which differs among the studyregions (Fig. 2). A pollen type was selected for use when it
Table 1
 Pollen-trap regions; PL Poland, CZ Czech Republic (Czech mountains), CH Switzerland (Swiss Alps), GE Georgia (Caucasus), LVLatvia, UK Great Britain (Wales)Country Pollen-trap region (abbr—full name) Latitude (
N) Longitude (
E) Elevation of traps (m a.s.l.) Pollen analyst
1
PL ROZ Roztocze 50.5812 23.0604 300
 ±
 50 IPCZ KRE E Krkonosˇe 50.73 15.77 1,150
 ±
 450 HS-SCZ KRW W Krkonosˇe 50.76 15.50 1,148
 ±
 348 HS-SCZ SUN N Sˇumava 49.09 13.40 1,095
 ±
 295 HS-SCZ SUS S Sˇumava 48.81 13.77 1,055
 ±
 320 HS-SCH GRI Grindelwald 46.6550 8.01 2,060
 ±
 335 JFNvLCH ALE Aletsch 46.3807 8.03 2,110
 ±
 110 JFNvLCH SIM Simplon 46.24 8.02 1,835
 ±
 250 JFNvLCH ZER Zermatt 45.00 7.78 2,620
 ±
 380 JFNvLGE LAG Lagodekhi 41.91 46.33 1,350
 ±
 950 EK, MCLV
2
LAT Latvia 56.70 24.37 95
 ±
 90 LUK 
2
CAP Capel Curig 53.09 3.90(
W) 320
 ±
 180 HP
1
All pollen analysts are co-authors
2
Data not used; explanation see text
Fig. 1
 Map showing pollen-trap countries. CH—Switzerland; CZ—Czech Republic; LV—Latvia; PL—Poland; FI—Finland; GE—Georgia; UK—Great Britain; see Table 1 for the regions within thecountriesVeget Hist Archaeobot (2010) 19:285307 287
 1 3
 
was encountered in all or most pollen traps of a region inall or most years of investigation.Climate dataThe climate data were obtained from the nearest availablemeteorological station(s) to the pollen-trap regions. Infor-mation on the climate data-sets employed is listed inTable 2. Figure 3 shows the average monthly temperature and precipitation for
 A
.
D
.1997–2007 of the climate data-setsconcerned. The climate data were not adjusted for altitudeaccordingtoregionallapserates, because (1)adjustmenthasno influence on the climate/pollen correlations made bylinear regression, (2) averages of the altitudes of the pollentraps are calculated by region, which would make adjust-ment somewhat arbitrary, and (3) the altitudes of the pollentraps can differ considerably from the altitude of the pollen-source populations, especially due to up-slope pollentransport from lowland taxa in mountain areas.We used climate data for the same parameters (tem-perature, precipitation) from two different stations for someregions when it was difficult to decide
 a priori
 whichstation would be more relevant to the pollen-source areasof the traps. In Poland, for example, one station (Zwie-rzyniec) is located in the slightly elevated study region of Roztocze, whereas the other station (Zamos´c´) lies in anearby, somewhat rain-sheltered valley. Also, the qualityof data may differ among stations, on which we have nofurther information.From the monthly values of the srcinal data-sets wecalculated multi-month averages of temperature and pre-cipitation sums comprising 2–6 consecutive months. Therationale of this is (a) that the pollen productivity of dif-ferent plant taxa may depend on weather conditions of different durations, and (b) single-month pollen/climatecorrelations are more prone to spurious statistical signifi-cance, especially for precipitation because it is spatiallymore variable than temperature. The climate time-serieswere linearly detrended.Statistical methodsThe pollen traps are grouped geographically in regions withthe same overall climate (Table 1). A common signal foreach pollen type in each region was extracted as follows. Aselection of traps was used in each region, choosing thelongest time-series with the fewest interruptions (Table 3).Therefore time-series of traps that had several years missingor were discontinued after some years due to various formsof disturbance, were not used. In each region, the medianyearly PAR of each pollen type was selected. The resultingtime-series (one per pollen type per region) were log-transformed and then linearly detrended. Each median PARvalue had 1 added, prior to log-transformation, to avoidproblems caused by the occurrence of zeros in some of themedian PAR time-series.The statistical software used is PAST v. 1.96 (Hammeret al. 2009). The pollen time-series were tested for normaldistribution, autocorrelation and outliers. PAR time-seriesare Poisson distributed rather than Gaussian, but logtransformation followed by detrending makes them close tonormally distributed. Autocorrelation in PAR time-series isto be expected, because a year of extremely high pollenproduction is often followed by a year with low pollenproduction for physiological rather than climatic reasons.However, after data transformation, no autocorrelation wasdetected statistically.Outliers in the pollen time-series might or might not berelated functionally to climate. We removed outliers toavoid ambiguity, on the basis of admittedly subjectivecriteria. Our aim was to create a balance between retainingpotential climate signals and removing potentially spuriousoutliers, as follows.(1) Outliers were statistically determined using PASTsoftware on log-transformed, detrended time-series.In addition, outliers that were visually detected innormal probability plots were removed to diminishthe chance of spurious results.(2) No more than one outlier was removed per pollentime-series, as outlier removal shortens the time-series and thus decreases the statistical strength of correlations. (One exception was made, shown inTable 4.)(3) Where there were several outliers in a pollen time-series, the most pronounced outlier was selected forremoval.(4) A statistically confirmed outlier was not removedwhen it lay close to another data point (as visuallydetermined on a normal probability plot). Thiscriterion retained pairs or clusters of outliers, which,because of their joined occurrence, are more likelyto be related functionally to weather conditions thanisolated outliers.Annual pollen values are compared to climate data inthe 20-month time window prior to pollen-trap collection(extending from January of the year before pollen-trapcollection up to, and including, August of the year of pollen-trap collection). Correlations were computedbetween single pollen types and single climate variables.The climate data considered are temperature and precipi-tation for all individual months and for all two-, three-,four-, five- and six-month periods falling within the 20-month time window. This results in 105 partly overlappingmulti-month periods per climate variable (temperature orprecipitation).
288 Veget Hist Archaeobot (2010) 19:285307
 1 3
 
The number of calculated correlations depends on thenumber of pollen types and climate variables per region. Itis well known that the number of spurious correlationsincreases with the number of statistical tests done. ABonferroni correction would reduce the
 P
-value to ca.0.0005 for a 5% chance that a correlation is spurious. Asthe aim of this paper is explorative rather than hypothesistesting, we show all correlations in Fig. 4 that have
P
\
0.05, but we are cautious with our interpretations.We refrain from providing proportions of explainedvariance. Presenting these would be over-rating the results,because pollen data are noisy, the calculation of the pollenvalues used in the correlations was complex, and thesuitability of the available climate data differs among thepollen-trap regions. Also, the explained variance for pre-cipitation can not be compared to that in temperaturebecause they have different relevance for the pollen-sourceareas.
Results and interpretation
The results for each pollen type are shown in Fig. 4.Results for temperature are in the upper section of eachdiagram, results for precipitation in the lower section. They-axis scale shows the months in the 20-month time win-dow prior to the month of pollen-trap collection, starting atthe top with 1
 =
 January. The horizontal dashed linesseparate 3-month seasons, the solid lines separate the years.The first header line indicates the study regions accordingto Table 1, the second header line indicates the climatestations according to Table 2, occasionally further abbre-viated due to space considerations. Correlations with
P
 C
 0.05 are not shown. Black dots show
 P
-values of significant correlations (
P
\
0.05), negative correlations tothe left of the central row of grey dots, positive to the right.The weakest correlations (
P
 close to 0.05) are placedclosest to the central line, the strongest (approaching
P
 =
 0) furthest away from it at the scale limit. For eachmonth, the correlation shown is the strongest (lowest
 P
-value) among the results of the overlapping multi-monthparameters (6–21 per month). There is one exception; theresult for a single-month parameter is not shown if its sign(
?
or
-
) is opposite to that for a multi-month parameter inthe same month, and no results are shown for months thathave both positive and negative results among the multi-month parameters. The number of months of the multi-month parameter used is printed to the right of the curves.
Table 2
 Climate data; the basic climate data used include average maximum monthly temperature (T) and monthly precipitation sums (P)Country
1
, region
1
, station Climate station Data source Parameter Latitude (N) Longitude (E) Elevation (m a.s.l.)PL ROZ ZWI Zwierzyniec BM
3
T
7
, P 50
8
37
0
22
8
58
0
230PL ROZ ZAM Zamos´c´ TuTiempo
4
T
7
, P 50
42
0
23
21
0
213CZ KRE PEC Pec pod Snezˇkou A.R.
5
T, P 50
40
0
15
45
0
820CZ KRE LIB, CZ KRW LIB Liberec A.R.
5
T, P 50
46
0
09
00
15
01
0
30
00
398CZ KRW HAR Harrachov A.R.
5
T, P 50
45
0
41.58
00
15
25
0
54.38
00
1,020CZ SUN CHU Chura´nˇov A.R.
5
T, P 49
04
0
06
00
13
36
0
47
00
1,118CZ SUN H&K Husinec
2
A.R.
5
P summer 49
03
0
17.85
00
13
59
0
13.1
00
492CZ SUS H&K Klatovy
2
A.R.
5
P winter 49
23
0
36
00
13
18
0
13
00
430CZ SUS CER Cˇerna´ v Posˇumavı´ A.R.
5
T 48
44
0
16.66
00
14
06
0
40.41
00
739CH GRI ABO Adelboden MeteoSwiss
6
T 46
29
0
31
00
7
33
0
40
00
1,320CH GRI GWD Grindelwald MeteoSwiss
6
P 46
37
0
8
02
0
1,158CH ALE EVO, CH SIM EVO Evole`ne-Villaz MeteoSwiss
6
T, P T 46
06
0
44
00
7
30
0
31
00
1,825CH SIM SIM Simplon-Dorf MeteoSwiss
6
P 46
11
0
41
00
8
03
0
17
00
1,495CH ZER ZER Zermatt MeteoSwiss
6
T, P 46
01
0
45
00
7
45
0
11
00
1,638GE LAG TBI Tblisi TuTiempo
4
T
8
41
40
0
44
55
0
467
1
Country and Region are abbreviated according to Table 1
2
Climate data of Husinec and Klatovy were combined together (H&K)
3
Co-author
4
5
Agrometeorological Rapporteur (1996–2008)
6
MeteoSchweiz, Zu¨rich CH
7
T 1997 of Zamosc was added to the time series of Zwierzyniec where was not measured in this year (BMK)
8
The P data have frequent gaps, and no other suitable P data from Georgia could be tracedVeget Hist Archaeobot (2010) 19:285307 289
 1 3
of 23