VARIATIONS OF THE CLIMATOLOGICAL GROWING SEASON ( ) IN GERMANY COMPARED WITH OTHER COUNTRIES

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 23: (2003) Published online in Wiley InterScience ( DOI: /joc.915 VARIATIONS OF THE CLIMATOLOGICAL GROWING SEASON ( ) IN GERMANY COMPARED WITH OTHER COUNTRIES ANNETTE MENZEL, a, * GERT JAKOBI, a REIN AHAS, b HELFRIED SCHEIFINGER c and NICOLE ESTRELLA a a Department of Ecology, Technical University of Munich, Germany b Institute of Geography, University of Tartu, Estonia c Central Institute for Meteorology and Geodynamics Vienna, Austria Received 9 August 2002 Revised 12 March 2003 Accepted 15 March 2003 ABSTRACT We analyse how variations in the climatological growing season, defined by single-value thresholds of daily minimum and mean air temperature, mirror recent changes in plant phenological phases. In Germany ( , 41 climate stations), the dates of last spring frost T min < 0 C (< 3 C and< 5 C) advance by 4 (3 and 0.32) days per year on average, and the first day when the daily mean temperature constantly exceeds 5 C (7 C and 10 C) advances by 0.13 (1 and 0.09) days per year. The respective autumn dates are delayed up to 5 days per year; in total, the climatological growing season is lengthened by 0.11 to 0.49 days per year, depending on the criterion analysed. However, a certain station-to-station variability is displayed. Mean trends of phenological phases correspond well to those measures of the climatological growing season having similar seasonal occurrence. Thus, for example, the extension of the growing seasons of trees (up to 2 days per year) averages the lengthening of the period where the daily mean temperature constantly exceeds 5 C (3 days per year). However, the greatest changes result for the lengthening of the frost-free period (0.49 days per year) due to the observed stronger increase in daily minimum rather than maximum temperatures. Consequently, trees may not take advantage of the frost-free season as before, but may profit from a reduced risk of damage by late spring frosts. A parallel evaluation of trends in the climatological growing season in other European countries (Austria, Switzerland, Estonia) supports this finding of a stronger lengthening of the frost-free period than that of the growing season based on daily mean air temperature. The mean lengthening of the frost-free period of 0.50 days per year in Austria and Switzerland ( , 18 stations <950 m a.s.l.) and 0.34 days per year in Estonia ( , two stations) is quite similar to that in Germany. The lengthening of the growing season (T mean 7 C) in Austria and Switzerland (0.39 days per year) also approximates the respective result for Germany (0.36 days per year). However, in several cases, the trends increase with altitude (up to m a.s.l.), but at high-elevation climate stations (>950 m a.s.l.) weaker trends are again generally observed. Copyright 2003 Royal Meteorological Society. KEY WORDS: climate change; growing season; frost-free period; phenology; air temperature; trend analysis; Germany 1. INTRODUCTION During the last few years an increasing number of studies have revealed a lengthening of the growing season across the Northern Hemisphere that is related to air temperature increases, and thus this is an important indicator of climate change (Keeling et al., 1996; Myneni et al., 1997; Magnuson et al., 2000; McCarthy et al., 2001; Menzel and Estrella, 2001; Tucker et al., 2001; Zhou et al., 2001; Walther et al., 2002; Root et al., 2003). This lengthening of the growing season can mostly be attributed to advances of spring phenology rather than the delay of autumn. Thus, primarily advances of spring leaf unfolding and flowering of plants are reported, ranging between days per decade in Europe and (3.8) days per decade in * Correspondence to: Annette Menzel, Lehrstuhl für Biometeorologie und Immissionsforschung, Am Hochanger 13, D Freising, Germany; menzel@met.forst.tu-muenchen.de Copyright 2003 Royal Meteorological Society

2 794 A. MENZEL ET AL. North America (Walther et al., 2002). Other proxy data for the overall seasonality, such as the ice cover of lakes and rivers (Magnuson et al., 2000; Sagarin and Micheli, 2001), or otherwise derived from the annual amplitude of mean temperature curves (Rapp and Schönwiese, 1994) or CO 2 concentrations (Keeling et al. 1996), as well as analyses of NOAA AVHRR NDVI satellite data (Myneni et al., 1997; Zhou et al., 2001; Tucker et al., 2001), confirm the results received from direct phenological plant observations (e.g. Menzel and Fabian, 1999; Menzel, 2000; Defila and Clot, 2001; Chmielewski and Rötzer, 2002). These changes and variations of the growing season length have important implications for the competition and fitness of plants, and thus may have profound ecological consequences (Walther et al., 2002). However, for many applications, such as the description of forest and agricultural sites or the integration of the length of the growing season in models, a plant-independent and more general definition is of interest. The variations of the start and end of this period are vital for many plants and insects, as both are particularly sensitive to the timing of extreme cold events at these times. Thus, our primary purpose is to examine variations in the climatological growing season, defined by common meteorological parameters, and compare them with the phenological growing season in Germany ( ). Additionally, results for the climatological growing season in Austria and Switzerland ( ) and Estonia ( ) are included to support the findings for Germany. Common climatological or meteorological definitions of the growing season often comprise the number of days between the last killing frost in spring and the first in the autumn (Brown, 1976). The killing temperature varies with plant species, stage of development, and the duration and rate of freezing; thus, the more general definition is the number of days between the last spring frost and the first autumn frost for different thresholds. Brinkmann (1979) found that long-term trends were highly sensitive to the definition of the growing season used in the first place. Therefore, in our study, we also analyse the growing season as the period between the first and last day where the daily mean temperature or the daily minimum temperature exceed various thresholds. Land air temperatures have increased by 0.6 ± C since the late 19th century, but the rate of increase for the night-time temperatures has been twice the rate of daytime temperatures since 1950, leading to a decreased diurnal temperature range (Karl et al., 1993; Easterling et al., 1997; McCarthy et al., 2001). The increase in the mean minimum temperatures has been demonstrated to have affected the length of the frost-free period (Karl and Easterling, 1999). Recent results, mainly for North America, reveal that the start of the frost-free season in the northeastern USA occurs 11 days earlier than in the 1950s (Cooter and LeDuc, 1995). Easterling et al. (2000) found that for the period there has been a slight decrease in the number of days below freezing over the entire USA, although with much regional variation in trends. Taking not the freeze-free period, but rather the period when the daily air temperature is maintained above 5 C, analyses of start, end, and length of the growing season from about 200 stations over the former Soviet Union indicated that little change has taken place over the last 110 years (Jones and Briffa, 1995). For northern and central Europe, Heino et al. (1999) found a decreasing number of frost days since the 1930s that is associated with strong increases in winter minimum temperatures. In Australia and New Zealand, the frequency of days below 0 C decreased with warming in daily minimum temperatures (Plummer et al., 1999). Robeson (2002) reported trends towards earlier spring frosts in Illinois ( ), but no consistent trends in autumn frosts. In total, the growing season has been lengthened by 1 week during the 20th century. Easterling et al. (2000) concluded that for every country where frost days have been examined (Australia, China, central and northern Europe, New Zealand, USA) they are becoming fewer in number. Bootsma (1994) reported for three stations in western Canada trends of earlier last spring frost, later first autumn frost, increased frost-free period, later growing season end and increased growing degree days, whereas two stations in eastern Canada did not exhibit the same warming trends. In his review, Robeson (2002) also summarized that nearly all of the more recent studies have reported increases in the growing season length. However, some studies show that, at individual stations, the link between the length of the growing season and the warmth during the growing season (often given as number of degree days above 5 C) is not intuitive and there may be little correlation (e.g. Bootsma, 1994; Jones and Briffa, 1995).

3 GERMAN GROWING SEASON DATA AND METHODS 2.1. Climate data A total of 41 climate stations of the German Weather Service (Deutscher Wetterdienst) were selected. These provided complete temperature records without any relocation of the station, good spatial coverage, and low potential for heat-island bias ( ). In addition, for an international comparison, temperature records of a further 20 Austrian and five Swiss climate stations ( ) and of two Estonian climate stations ( ) were used (Figure 1). All daily minimum T min and mean air temperature T mean data were carefully checked, but none of the small gaps was filled. The time series were tested visually in different plots to exclude larger breaks in homogeneity. However, the temperature series were not homogenized. Only at the climate station 2290 (Pirmasens, Germany) was there a strong bias in T min during of unknown origin, which had to be adjusted Phenological data To compare changes of the climatological growing season with the phenological growing season, we used the recently constructed phenological anomalies ( ) from the phenological network of the German Weather Service (Menzel, 2003). Some 16 phenological phases of wild plants that cover the entire growing season were selected; in addition, the growing season length, defined as the time between leaf unfolding and colouring of four deciduous tree species, was determined. Figure 1. Climate stations in Germany (1), Austria (2), Switzerland (3) and Estonia (4) selected for analysis

4 796 A. MENZEL ET AL. Table I. Measures for length, start and end of the climatological growing season Measures based on T min (daily minimum air temperature) Measures based on T mean (daily mean air temperature) Spring LSF T min < 0 C LSF T min < 3 C LSF T min < 5 C Autumn FFF T min < 0 C FFF T min < 3 C FFF T min < 5 C Season FFP T min < 0 C FFP T min < 3 C FFP T min < 5 C Last spring frost with T min below the given threshold FD T mean 5 C FD T mean 7 C FD T mean 10 C First day on which T mean constantly equals or exceeds the given threshold First fall frost with T min below the given threshold LD T mean 5 C LD T mean 7 C LD T mean 10 C Last day of the period in which T mean constantly equals or exceeds threshold Frost-free period with no case of T min below the threshold GS T mean 5 C GS T mean 7 C GS T mean 10 C Growing season in which on all days T mean is equal to or above threshold 2.3. Methods Out of the different possible definitions of the climatological growing season, we selected the number of days between the last and first occurrence of minimum temperatures T min < 0, 3, and 5 C. These three temperature thresholds were used to determine, for each climate station and year, the dates of the last spring frost (LSF), the first fall frost (FFF) and the frost-free period (FFP). The two lower thresholds represent progressively harder frosts; the threshold of 5 C is a useful indicator for cold-sensitive plants (see also Robeson (2002)). In addition, the growing season (GS) as the period in which T mean constantly equals or exceeds the thresholds T mean 5, 7, and 10 C was determined. Short drops of T mean below these limits in June, July and August were neglected. The start of the growing season is then defined by the first day (FD) in spring on which T mean constantly equals or exceeds the respective threshold, and the end by the last day (LD) in autumn (see Table I). Apart from the different temperatures used (T min and T mean ), LSF and FD (as well as FFF and LD) would only differ by 1 day for the same temperature threshold. The beginning of the xylem growth, after von Wilpert (1990), was then used as supplementary information for the start of the growing season in Germany. For each climate station and year, this was determined as the average of (1) the date on which the 7 day running mean air temperature exceeds 10 C for the first time and (2) the date on which it exceeds 10 C for at least five consecutive days. For each station, all time series of the frost-free period (LSF, FFF, FFP, T min < 0, 3, 5 C) and the growing season (FD, LD, GS T mean 5, 7, 10 C) were regressed against time (significance after F -test p<0.05, p<0.01, p<0.001). Hereafter, if no level of significance is specified, then all trends significant at the most liberal level (p <0.05) are referred to as significant. Linear regressions of the resulting trends on longitude, latitude, and altitude were calculated to investigate geographical effects. To study mean conditions in Germany, the results of the individual 41 climate stations were averaged. The time series of the climatological seasons were also recombined to mean anomalies, which were determined, for each climate station and all growing season measures of Table I, as their annual deviations from the respective station average. Pearson product-moment correlations of the climatological and meteorological anomalies were calculated to analyse the temporal coherence of the different climatological measures of the growing season and the onset of phenological phases Trends of the climatological growing season 3. RESULTS Single station time series of the different measures of the climatological growing season were relatively noisy, but distinct trends were present at many stations. For all thresholds, the linear trends of the start, end

5 GERMAN GROWING SEASON 797 and length of the frost-free period, as well as of the growing season, have been tabulated, e.g. for the threshold T min < 0 C in Table II and for T mean 7 C in Table III. In Germany, except for the two climate stations Pirmasens (2290) and Eutin (3822), an earlier onset of the last spring frost by 0.05 to 0.56 days per year, a later onset of the first fall frost by 0.04 to 0.55 days per year and a lengthening of the frost-free period by 0.01 to 1.07 days per year is revealed (Table II). Although minimum temperatures at climate station 2290 have already been corrected, this climate station often shows opposite results, such as a shortening of the frost-free period for all three thresholds. At 63% of the German climate stations the trend of FFP is significant; at 44% of the climate stations the trends of LSF and FFF are significant. Our additional analysis for the Austrian and Swiss climate stations (<950 m a.s.l.) supports these findings, with an observed earlier onset of the frost-free period by 0.02 to 0.57 days per year, a later onset of the first fall frost by 0.10 to 0.60 days per year and a lengthening of the frost-free period by 0.09 to 1.16 days per year, leading to similar mean values for the start, end and length of the frost-free period. However, all respective trends for the high-elevation climate stations (>950 m a.s.l.) are lower; thus the changes in the frost-free period are weaker. In Estonia, the two climate stations analysed also reveal a lengthening of the frost-free period by 0 and 0.47 days per year. For the threshold T mean 7 C (Table III), all stations in Germany, except Pirmasens (2290), have experienced increases in the growing season duration by 0.03 to 0.76 days per year. Nearly all stations show trends towards an earlier first date (0.05 to 0.39 days per year), when the temperature threshold of 7 C is constantly exceeded. Nearly all of them also reveal trends towards later dates in autumn ( 0.05 to 0.32 days per year), when the temperature first drops below that threshold. The range of the lengthening of the growing season (T mean 7 C) in Austria and Switzerland (stations <950 m a.s.l.) is nearly identical (0.14 to 0.78 days per year), with a comparable earlier start ( 0.01 to 0.53 days per year) and later end ( 0.16 to 0.34 days per year). At stations above 950 m the observed changes are generally smaller. For two Estonian climate stations, both FD and LD occur earlier; thus, almost no changes in the length of the growing season are observed Spatial variation of the climatological trends Figures 2 to 4 comprise all the trends of the different measures of the climatological growing season (Table I) in Germany and show, at the same time, their spatial variation. In general, the start of spring advances, especially for LSF dates and FD T mean 7 C (Figure 2). Except for LD T mean 10 C, the timing of autumn events is delayed, in particular for the southern and eastern parts of Germany (Figure 3). In total, the growing season and the frost-free period are lengthening (Figure 4). As with the recently published maps of phenological trends in Germany (Menzel and Estrella 2001; Menzel et al., 2001), there is a certain spatial variability displayed for the trends of the 18 climatological measures (Figure 2 4). However, in contrast to phenological changes, these shifts are partly found to depend on geographical parameters in Germany. Out of all regressions of trends (18 different measures; see Table I) against latitude, longitude and altitude (not displayed here), only 35% are significant, which at the same time possess notable proportions of the common variance explained (between 10 and 42%). In Germany, altitude is related to latitude, and thus altitudinal and latitudinal effects are strongly correlated. Both spring advance and autumn delay are more pronounced at lower latitudes and higher altitudes than in the other parts of the country (T mean 5 C andt min < 0 C). In addition, the trends of FFF T min < 3 C are also higher at lower latitudes. Relevant longitudinal effects are revealed for GS and LD T mean 7 C, with a clearer delay of LD and a stronger lengthening of the GS in the eastern part of Germany. However, taking the additional results of the Austrian, Swiss and Estonian climate stations into account, we find, as Figure 5(a) displays for T mean 7 C and Figure 5(b) for T min < 0 C, that the relationship of trends against latitude is interrupted in the Alpine region, where altitude and latitude are decoupled. All plots of trends against altitude, however, show that the changes (lengthening of the FFP, advance of LSF and delay of FFF) are most pronounced at medium-elevation stations (around 500 m a.s.l. in southern Germany and parts of Austria and Switzerland), where they reach a maximum. Trends are smaller at low elevations

6 798 A. MENZEL ET AL. Table II. Linear trend (days per year) of the length of frost-free period (FFP), of the last spring frost (LSF) and the first fall frost (FFF) (T min < 0 C), N number of years, F -test, significance level and P -value Station ID Lat Lon Alt N FFP trend LSF trend FFF trend F -test Significance P -value F -test Signi- P -value F -test Signi- a ficance a ficance a P -value Germany Ückermünde D ** ** 0.00 Boltenhagen D Nordhorn D * ** Schleswig D Boizenburg D Teterow D Gardelegen D Eutin D Berlin- D * Dahlem Angermünde D ** * Schwerin D Cottbus D * ** Magdeburg D Potsdam D * * Doberlug- D * Kirchhain Lindenberg D Wertheim- D ** ** 0.01 Eichel Leipzig D Artern D * Göttingen D ** * * 0.05 Saarbrücken- D *** *** *** 0.00 St. Jo. Kassel D Görlitz D * * Mergentheim D ** ** 0.01 Trier- D *** * ** 0.00 Petrisberg Erlangen D *** * *** 0.00 Pirmasens D Bad D Königshofen Hilgenroth D ** * * 0.01 Biedenkopf D * Kronach D * * 0.01 Metten D *** ** *** 0.00 Höllenstein- D ** ** 0.01 Kraftw. Chemnitz D *** ** * 0.02 Weissenburg/ D ** ** 0.00 BY Ellwangen/ D *** *** *** 0.00 Jagst Wörnitz- D *** ** ** 0.01 Bottenw. Deuselbach D

7 GERMAN GROWING SEASON 799 Table II. (Continued) Station ID Lat Lon Alt N FFP trend LSF trend FFF trend F -test Significance P -value F -test Signi- P -value F -test Signi- a ficance a ficance a P -value Hof-Hohensaas D *** *** *** 0.00 Schwangau-Horn D *** *** *** 0.00 Hohenpeissenberg D ** * * 0.03 Austria/ Switzerland Aspang A *** *** *** 0.00 Bad Ragaz CH ** ** * 0.03 Basel-Binningen CH *** ** ** 0.01 Graz-Flughafen A *** *** ** 0.00 Hoersching A * ** 0.00 Kollerschlag A Kremsmuenster A Kufstein A Luzern CH Muerzzuschlag A *** *** ** 0.00 Oberleis A Salzb-Flughafen A *** ** * 0.02 Schoppernau A Schwechat A Waizenkirchen A *** *** ** 0.00 Wien- A Hohe Warte Woerterberg A Zuerich SMA CH ** ** * 0.05 >950 m a.s.l. Villacher-Alpe A Mooserboden A Feuerkogel A Davos CH Stolzalpe A Holzgau A Kornat A Estonia Tartu EST *** *** 0.00 Pärnu EST ** a *5% level; **1% level; ***0.1% level. (northern Germany) and in higher mountainous stations (>950 m a.s.l.). None of the measures shown in Figure 5 clearly depends on longitude (not displayed) Comparison of climatological trends with phenological changes The summarizing statistics of the trends of the climatological growing season in Germany (mean, maximum and minimum trend and percentage of stations with significant trends; Table IV) clearly reveal that the results are sensitive to the particular definition used. The lengthening of the growing season is most apparent for the frost-free period, especially for T min < 0 C with 0.49 days per year exceeding the mean values for T min < 3 C (0.40 days per year) and T min < 5 C (0.39 days per year). The growing season also tends to become longer (by 0.36 days per decade for T mean 7 C; less for T mean 5and 10 C). Spring measures show clear tendencies towards an earlier start especially for all LSF and FD T mean 7 C (between 1 and 0.32 days per decade). Autumn measures, on average, tend to occur later, but these shifts are smaller than

8 800 A. MENZEL ET AL. for the respective changes in spring. Only for FFF T min < 0 C do the autumn changes equal those in spring, with a maximum delay of 5 days per year. The percentage of climate stations with significant trends varies among thresholds and seasons, ranging between 66% (63%) for GS T mean 7 C (FFP T min < 0 C respectively) and 5% for LD T mean 10 C and FFF T min < 5 C. Table III. Linear trends (days per year) of the growing season length (GS), of the first day in spring (FD) and the last day in autumn (LD) (T mean 7 C), N number of years, F -test, significance level and P -value Station ID Lat Lon Alt N GS trend FD trend LD trend F -test Significance P -value F -test Signi- P -value F -test Signi- a ficance a ficance a P -value Germany Ückermünde D * Boltenhagen D * Nordhorn D * * Schleswig D Boizenburg D * Teterow D ** * Gardelegen D Eutin D Berlin-Dahlem D * Angermünde D ** * * 0.03 Schwerin D * * 0.04 Cottbus D * * Magdeburg D ** * 0.02 Potsdam D *** ** * 0.04 Doberlug-Kirchhain D *** *** Lindenberg D * Wertheim-Eichel D ** 0.01 Leipzig D ** * 0.04 Artern D ** * Göttingen D Saarbrücken-St. Jo. D * 0.02 Kassel D Görlitz D ** ** Mergentheim D Trier-Petrisberg D Erlangen D *** ** 0.00 Pirmasens D * 0.02 Bad Königshofen D * * 0.03 Hilgenroth D * ** Biedenkopf D * Kronach D *** * ** 0.00 Metten D * * 0.03 Höllenstein-Kraftw. D * ** Chemnitz D * Weissenburg/BY D *** *** * 0.04 Ellwangen/Jagst D *** ** * 0.02 Wörnitz-Bottenw. D *** ** ** 0.00 Deuselbach D * Hof-Hohensaas D Schwangau-Horn D * *** Hohenpeissenberg D Austria/Switzerland Aspang A *** *** Bad Ragaz CH ** *

9 GERMAN GROWING SEASON 801 Table III. (Continued) Station ID Lat Lon Alt N GS trend FD trend LD trend F -test Significance P -value F -test Signi- P -value F -test Signi- a ficance a ficance a P -value Basel-Binningen CH * 0.02 Graz-Flughafen A *** ** ** 0.00 Hoersching A * 0.04 Kollerschlag A ** Kremsmuenster A * ** 0.00 Kufstein A *** *** * 0.02 Luzern CH Muerzzuschlag A * ** Oberleis A Salzb-Flughafen A *** *** * 0.02 Schoppernau A *** Schwechat A * * 0.02 Waizenkirchen A * * Wien-Hohe Warte A Woerterberg A Zuerich SMA CH >950 m a.s.l. Villacher-Alpe A Mooserboden A Feuerkogel A Davos CH Stolzalpe A * Holzgau A ** Kornat A * Estonia Tartu EST *** Pärnu EST a *5% level; **1% level; ***0.1% level. On comparing the mean trends of start, end and length of the climatological growing season at the 41 stations in Germany with the mean phenological trends (Menzel, 2003) for the period (Figure 6), it turns out that phenological and climatological changes correspond quite well in their seasonal occurrence. For example, the trend of the start of the frost-free period, especially for T min < 3 C and< 5 C, corresponds to phases of the earliest spring, such as flowering of snowdrops and forsythia. The advance of the first day T mean 5 C mirrors the changes in leaf unfolding of deciduous trees and flowering of fruit trees, and the trend of growing season GS T mean 5 C also matches the changes of the phenological growing season best. However, the advance of the frost-free period exceeds most of the phenological changes, such as advances of leaf unfolding and flowering of trees. Thus, the intensity of plant reaction in autumn and late spring is smaller than that of pure meteorological factors based on frost events. This means that the potential growing season, as defined by the FFP, may not be exploited to the same extent as in the past. At the same time, the potential risk of frost damage in spring may decrease due to this divergent intensity of daily minimum temperature and phenological changes Temporal coherence of phenological and climatological growing season Figure 7 shows that the anomalies (annual deviations from the average) of the frost-free period and the phenological growing season of deciduous trees in Germany are not necessarily identical. Thus, the temporal coherence of phenological and climatological growing season in the second half of the 20th century is not as good as the comparison of their mean trends suggests.

10 802 A. MENZEL ET AL. Figure 2. Linear trends of different spring parameters of the climatological growing season. (A) FD first day T mean 10 C, (B) FD T mean 7 C, (C) FD T mean 5 C, (D) LSF T min < 0 C, (E) LSF T min < 3 C, (F) LSF T min < 5 C (for further description see Table I). This figure is available in colour online at We determined the correlation coefficients of all possible pairs of phenological phases and the climatological measures of the growing season. In Table V, however, we only summarize the highest correlation coefficients between phenological and climatological anomalies. The anomalies of succeeding phenological phases in the seasonal cycle correspond quite well (correlation coefficients between 0.84 and 0.98). Flowering phases in spring correlate best with flowering of other species, and the length of the growing season is linked to each species leaf unfolding in spring. The correlation coefficients for the best correlating biotic variable are, in every case, higher than for the best abiotic variables.

11 GERMAN GROWING SEASON 803 Figure 3. Linear trends of different autumn parameters of the climatological growing season. (A) LD T mean 10 C, (B) LD T mean 7 C, (C) LD T mean 5 C, (D) FFF T min < 0 C, (E) FFF T min < 3 C, (F) FFF T min < 5 C (for further description see Table I). This figure is available in colour online at Of the abiotic variables, LSF T min < 5 C and the beginning of the xylem growth after von Wilpert (1990) correlate best with spring phases and the length of the growing season, and the first day T mean 7 C or 10 C correlate best with anomalies of summer and autumn phases. Nevertheless, the anomalies of the climatological growing season, which is defined by singular events, are only linked to a certain extent to plant reactions, which integrate the weather of preceding months. Thus, for both spring phases and the length of the growing season, the beginning of xylem growth after von Wilpert (1990), based on running temperature means, often surpasses the correlations with measures of the start of the climatological growing season.

12 804 A. MENZEL ET AL. Figure 4. Linear trends of different measures of the climatological growing season. (A) GS T mean 10 C, (B) GS T mean 7 C, (C) GS T mean 5 C, (D) FFP T min < 0 C, (E) FFP T min < 3 C, (F) FFP T min < 5 C (for further description see Table I). This figure is available in colour online at Lengthening of the growing season 4. DISCUSSION The increase in the global mean temperature of about 0.6 C since the start of the 20th century is associated with a stronger warming in daily minimum temperatures than in maximum temperatures, this asymmetry is detectable in all seasons and in most of the regions studied (McCarthy et al., 2001). Between 1951 and 1990, the rise in minimum temperature has even occurred at a rate three times that of the maximum temperature (0.84 C versus8 C) for 50% (10%) of the Northern (Southern) Hemisphere landmass (Karl et al., 1993).

13 GERMAN GROWING SEASON 805 a LSF Germany Austria, Switzerland >950 m a.s.l. Estonia FFF FFP latitude altitude Figure 5. Trends (days per year) of (a) the last spring frost (LSF), first fall frost (FFF) and length of frost-free period (FFP) T min < 0 C and (b) the first day (FD), last day (LD) and the length of the growing season (GS) T mean 7 C against latitude ( N) and altitude (m a.s.l) In association with this temperature increase there is a lengthening of the growing season apparent in the mid and higher latitudes of the Northern Hemisphere from different data sets, such as satellite images, CO 2 records and phenological ground truth. However, the temperature records themselves also reveal the lengthening of the growing season in the last few decades. Most of the numerous existing definitions of the climatological growing season use the dates when air temperature exceeds a threshold in spring and falls below it in autumn. Often, the resulting time series are very noisy (Robeson, 2002), especially if single-value thresholds are used instead of multiple-day criteria or running temperature means. Easterling et al. (2000) pointed out that one of the major problems in examining the climate records for changes in extreme events is a lack of high-quality, long-term data. In our study, the careful data check revealed (excluding problems encountered at climate station 2290) that the temperature data were of high quality. Owing to the demand of complete meteorological and phenological data coverage, our study only comprises the second half of the 20th century ( ). However, in much of Europe, the recent decades have been approximately the warmest of the instrumental period, but there were also warmer periods in the 1930s and some decades in the

14 806 A. MENZEL ET AL. b FD Germany Austria, Switzerland >950 m a.s.l. Estonia LD GS latitude altitude Figure 5. (Continued) period in Scandinavia and part of central Europe (Watson et al., 1997; Yan et al., 2002). Longterm temperature variability in the European Alps ( ) includes an initial decrease of the annual and seasonal series to a minimum followed by a positive trend until 1998 (Böhm et al., 2001). Focusing on extreme climate changes over the last years, Yan et al. (2002) found decreasing warm extremes up to the late 19th century, decreasing cold extremes since then and increasing warm extremes since To determine the climatological growing season, we used six single-value thresholds, based on T min and T mean, and the beginning of xylem growth after von Wilpert (1990), which represents a multiday criterion. On average, for Germany, Austria, Switzerland, and Estonia, we found an earlier start and a later end of the climatological growing season. However, as described above, long-term fluctuations in T mean and T min are not necessarily the same, and temperatures trends are not necessarily the same for different months. Thus, the resulting trends of the climatological growing season by different measures clearly vary, e.g. weaker trends for the growing season T mean 7 C in Estonia Effects of the asymmetric diurnal temperature change Studies analysing several plant species often report a strong seasonal variation of changes, with strongest advances of the earliest spring phases (Menzel et al., 2001). Thus, a comparison of changes derived by the

15 GERMAN GROWING SEASON Galanthus nivalis F Forsythia F Betula pendula LU Aesculus hippocastanum LU 0.4 Prunus avium F Fagus sylvatica LU Malus domestica F Quercus robur LU Picea abies LU Tilia platyphyllos F Sambucus nigra FR Aesculus hippocastanum FR Aesculus hippocastanum LC Betula pendula LC Fagus sylvatica LC Quercus robur LC Aesculus hippocastanum GS Quercus robur GS Betula pendula GS Fagus sylvatica GS LSF Tmin<-5 C LSF Tmin<-3 C FFF Tmin<-5 C FFF Tmin<-3 C FFP Tmin<-5 C FFP Tmin<-3 C LSF Tmin<0 C FD Tmean> = 5 C Mean trend [days / year] FFF Tmin<0 C LD Tmean> = 5 C FFP Tmin<0 C GS Tmean> = 5 C -0.3 FD Tmean> = 7 C FD Tmean> = 10 C LD Tmean> = 7 C LD Tmean> = 10 C GS Tmean> = 7 C GS Tmean> = 10 C Mean onset / duration [day since Jan 1st / days] Figure 6. Mean trends of phenological phases and measures of the climatological growing season against their mean onset in Germany ( ). Trends of the growing season on/at the right margin, phenological phases (F flowering, LU leaf unfolding, FR fruit ripening, LC leaf colouring, GS growing season) from Menzel (2003), definitions of the climatological growing season in Table I

16 808 A. MENZEL ET AL. Table IV. Statistics of the trends of the climatological growing season for 41 climate stations in Germany: mean, maximum and minimum days per year; number of stations N with significant trends at the 5%, 1% and 0.1% levels; percentage of climate stations with significant trends; absolute change for Days per year N Stations Change Mean Max Min 5% level 1% level 0.1% level significant (%) (days) T mean 10 C GS FD LD T mean 7 C GS FD LD T mean 5 C GS FD LD T min < 0 C FFP LSF FFF T min < 0 C a FFP LSF FFF T min < 3 C FFP LSF FFF T min < 5 C FFP LSF FFF a Excluding stations 2290 and Anomaly [days] FFP Tmin<0 C FFP Tmin<-3 C FFP Tmin<-5 C GS deciduous trees Figure 7. Mean annual anomalies of the frost-free period (FFP) and of the mean length of the growing season (B. pendula, A. hippocastanum, F. sylvatica, Q. robur; Menzel, 2003) in Germany ( )

17 GERMAN GROWING SEASON 809 Table V. Coherence of annual anomalies in Germany. Correlation coefficients (r, bold p<0.001, normal p<0.01) between phenological anomalies (Menzel, 2003) and the two best biotic variables (phenological anomalies, F flowering, LU leaf unfolding, FR fruit ripening, LC leaf colouring, GS growing season), and best abiotic variable (climatological measures, see Table I, Wilpert beginning of xylem growth after von Wilpert (1990)) Phase Best biotic variable r 2nd best biotic variable r Best abiotic variable r Galanthus nivalis F F. suspensa F 0.88 M. domestica F 0.69 LSF T min < 5 C 0.69 Forsythia suspensa F G. nivalis F 0.88 P. avium F 0.86 LSF T min < 5 C 0.68 Aesculus B. pendula LU 0.98 P. avium F 0.95 Wilpert 0.72 hippocastanum LU Betula pendula LU A. hippocastanum LU 0.98 P. avium F 0.96 Wilpert 0.74 Fagus sylvatica LU Q. robur LU 0.95 M. domestica F 0.93 Wilpert 0.71 Quercus robur LU P. abies LU 0.96 F. sylvatica LU 0.95 Wilpert 0.77 Picea abies LU Q. robur LU 0.96 M. domestica F 0.95 Wilpert 0.71 Prunus avium F B. pendula LU 0.96 M. domestica F 0.95 Wilpert 0.70 Malus domestica F P. abies LU 0.95 P. avium F 0.95 Wilpert 0.73 Tilia platyphyllos F A. hippocastanum FR 0.95 S. nigra FR 0.95 FD T mean 7 C 0.76 Sambucus nigra FR T. platyphyllos F 0.95 A. hippocastanum FR 0.91 FD T mean 7 C 0.75 Aesculus T. platyphyllos F 0.95 S. nigra FR 0.91 FD T mean 7 C 0.71 hippocastanum FR Aesculus F. sylvatica LC 0.84 A. hippocastanum FR 0.79 FD T mean 7 C 0.45 hippocastanum LC Fagus sylvatica LC Q. robur LC 0.92 B. pendula LC 0.88 LD T mean 10 C 0.51 Quercus robur LC F. sylvatica LC 0.92 B. pendula LC 0.83 LD T mean 10 C 0.43 Betula pendula LC F. sylvatica LC 0.88 Q. robur LC 0.83 LD T mean 10 C 0.38 Aesculus B. pendula GS 0.95 A. hippocastanum LU 0.90 Wilpert 0.62 hippocastanum GS Betula pendula GS A. hippocastanum GS 0.95 B. pendula LU 0.93 Wilpert 0.66 Fagus sylvatica GS Q. robur GS 0.92 A. hippocastanum GS 0.90 LSF T min < 5 C 0.59 Quercus robur GS F. sylvatica GS 0.92 B. pendula GS 0.86 LSF T min < 5 C 0.58 different methods should only be made by taking their mean seasonal occurrence into account: in spring, LSF T min < 0 C andfdt mean 5 C occur almost on the same date on average in Germany (day 115/116 since 1 January , 41 climate stations); also, FFF T min < 0 C andldt mean 5 C (day 293/296) in autumn. However, the advance of spring is greater for LSF ( 4 days per year) than for FD ( 0.13 days per year) and the delay of autumn greater for FFF (5 days per year) than for LD (0.09 days per year), leading to a stronger lengthening of the frost-free period T min < 0 C (0.49 days per year) than of GS T mean 5 C (3 days per year). Here, we have strong evidence that the observed asymmetric diurnal temperature change is reflected by a stronger lengthening of the frost-free period than of the almost simultaneous growing season constantly exceeding 5 C Uncertainties Some related questions are not yet solved: Brinkmann (1979) noted that T min is particularly sensitive to small local differences and is characteristically a much more variable element in its horizontal distribution than daily mean temperatures. Although mean changes in the climatological growing season, derived from 41 climate stations, may give representative results for Germany, their spatial resolution is not sufficient to compare the spatial variability of changes based on T min and T mean. Growing-season statistics based on a single-threshold air temperature can sometimes be misleading (Brinkman, 1979). Therefore, we used three thresholds for both T min and T mean. The resulting advance of spring and delay of autumn in Germany based on T min are nearly in consecutive order, whereas the

18 810 A. MENZEL ET AL. respective findings for T mean 7 C clearly deviate from those for the adjacent thresholds. This apparent result may be caused, apart from different temperature trends in different months, by variations due to single-value criteria Comparison of the climatological and phenological growing season Our primary research focus was the comparison of changes of the climatological and phenological growing season. Both for the advance of spring and for the delay of autumn, the observed mean trends in Germany match well when climatological criteria and phenological phases of similar seasonal occurrence are compared. The lengthening of phenological growing of trees from leaf unfolding to leaf colouring is mirrored by the climatological growing season GS T mean 5 C. Thus, our study confirms that the increase in mean temperature during recent decades has lengthened the foliation period of trees. However, for Germany, Austria and Switzerland (stations below 950 m a.s.l.), greater changes have been revealed for the frost-free period FFP T min < 0 C, probably due to the greater increase in daily minimum temperatures. This asymmetric diurnal temperature change is not observed at the mountain-top stations in the Alpine region (Weber et al., 1994; Rebetez and Beniston, 1998). This fact might account for the smaller lengthening of the frost-free period above 950 m a.s.l compared with below 950 m that we found for Swiss and Austrian stations. In general, our findings suggest that, today, trees do not profit to the same extent from the frost-free period as half a century ago and that, on average, the potential risk of damage by late spring frost for the new leaves and flowers of the trees may have decreased (see also Scheifinger et al. (2003)). A similar result was also found by a phenological and photosynthetic model comparison by Linkosalo (2001) for the last 93 years in Finland. He concluded that, owing to relatively late bud burst, compared with frost damage thresholds, two common boreal tree species lose up to 15% of their potential photosynthetic production. We plan further studies to assess the changing risk of frost damage at single sites in Germany and to verify these preliminary results. Unfortunately, phenological observations rarely go beyond 1000 m (in a very few cases they go up to 1350 m in the Alpine region) because they depend strongly on (permanent) human settlements. Thus, we cannot evaluate if the smaller lengthening of the frost-free period above 950 m is connected with smaller changes of the phenological growing season there. Nevertheless, our findings of an altitudinal gradient in central Europe with maximum changes of the climatological growing season between 500 and 950 m a.s.l. is well mirrored by phenological observations in Switzerland (Defila and Clot, 2001), where the proportion of advancing trends is predominantly increasing with altitude from 500 to 1100 m. Although mean trends of the measures of the phenological and climatological growing season, defined by single-value thresholds, match quite well, their temporal coherence is not as good as those of succeeding phenological phases. The use of a 5 day running temperature by von Wilpert (1990) has a smoothing and balancing effect: the mean time series of the beginning of xylem growth corresponds better to phenological anomalies of late spring than the start of the climatological growing season. Thus, if not only mean trends, but also time series of phenological phases are to be approximated from climatological data, then it is recommended not to rely on single-value thresholds, but on integrating, smoothing measures (e.g. Bootsma, 1994). ACKNOWLEDGMENTS We kindly thank the Deutscher Wetterdienst (German Weather Service), Swiss Meteo, the Central Institute for Meteorology and Geodynamics, and the Estonian Meteorological and Hydrological Institute for providing meteorological and phenological data, and Jared David May for looking over the language. This study was carried out within the project POSITIVE, which was financed by the European Commission within FP5 (EVK2-CT ).