Effects of recent warm and cold spells on European plant phenology
- First Online:
- Cite this article as:
- Menzel, A., Seifert, H. & Estrella, N. Int J Biometeorol (2011) 55: 921. doi:10.1007/s00484-011-0466-x
- 340 Views
Climate change is already altering the magnitude and/or frequency of extreme events which will in turn affect plant fitness more than any change in the average. Although the fingerprint of anthropogenic warming in recent phenological records is well understood, the impacts of extreme events have been largely neglected. Thus, the temperature response of European phenological records to warm and cold spells was studied using the COST725 database. We restricted our analysis to the period 1951–2004 due to better spatial coverage. Warm and cold spells were identified using monthly mean ENSEMBLES temperature data on a 0.5° grid for Europe. Their phenological impact was assessed as anomalies from maps displaying mean onsets for 1930–1939. Our results clearly exhibit continental cold spells predominating in the period 1951–1988, especially during the growing season, whereas the period from 1989 onwards was mainly characterised by warm spells in all seasons. The impacts of these warm/cold spells on the onset of phenological seasons differed strongly depending on species, phase and timing. “False” phases such as the sowing of winter cereals hardly reacted to summer warm/cold spells; only the sowing of summer cereals mirrored spring temperature warm/cold spells. The heading dates of winter cereals did not reveal any consistent results probably due to fewer warm/cold spells identified in the relevant late spring months. Apple flowering and the harvest of winter cereals were the best indicators of warm/cold spells in early spring and summer, also being spatially coherent with the patterns of warm/cold spells.
KeywordsClimate changeExtreme eventsHeat wavesTemperature response
Existing or projected consequences of climate change include alterations in the frequency, intensity, geographic scale, and location of extreme weather and climate events (e.g. Horton et al. 2001, Trenberth et al. 2007). For many phenomena, such as warmer and more frequent hot days on land, trends in the twentieth century were significant, the human contribution is likely and continued trends for the twenty-first century are virtually certain (Parry et al. 2007). Although climate change is strongly linked to more and stronger physical extreme events, not all will necessarily translate into extreme impacts since systems may be naturally resilient or well adapted by management or special adaptation measures (e.g. Easterling et al. 2000; Parmesan et al. 2000; Huynen et al. 2001; Chau et al. 2009). Consequently, the perspectives on extreme events vary broadly, from statistical definitions of measured physical attributes of phenomena used by climatologists (e.g. Easterling et al. 2000; Kioutsioukis et al. 2010; Katz 2010) to impact related approaches in natural sciences (e.g. Parry et al. 2007). Sometimes, both aspects are linked if an extreme event is seen as a notable, rare or unique event or otherwise significant, e.g. in terms of its impacts.
Extremes are commonly defined on the basis of the tails of distributions (e.g. the upper or lower 10th percentiles; Beniston 2009), by exceedance of absolute thresholds (e.g. Satyamurty et al. 2007) or by measures of variation (e.g. Jones et al. 1999; Horton et al. 2001; Katz 2010). Warm and cold spells are not single extreme events, but can be regarded as a compound extreme, i.e. as a persistence of weather conditions, comparable to drought. However, although the term “warm and cold spells” is frequently used (e.g. Huynen et al. 2001; Beniston 2005), there is no general definition of it. Spells and their frequency, duration, or intensity are defined variously in the literature (Jones et al. 1999; Horton et al. 2001; Shabbar and Bonsal 2003; Satyamurty et al. 2007; Beniston 2009; Klein Tank et al. 2009; Katz 2010).
Periods of at least six consecutive days with daily mean temperatures exceeding either a percentile of the distribution or a baseline temperature have been suggested (e.g. Klein Tank et al. 2009), while other definitions use mean durations of warm/cold spells longer than 11 days (e.g. Satyamurty et al. 2007). In general, the concept of using relative temperature thresholds instead of absolute ones seems widely accepted (e.g. McCalla et al. 1978).
In the past, climate change-driven variations in extreme events were mainly studied in physical systems, and their impacts on the natural environment were mostly neglected. Recent research suggests that extreme events have been and, in the course of accelerated anthropogenic climate change, will be the major driving factor influencing fitness (growth and survival), and thus distributional ranges and sustainability of ecosystem services. In particular, the effects of winter warm spells on dehardening and frost resistance have been studied. However, the observed responses differed with duration, timing and species, ranging from a dramatic decrease in frost hardiness, with or without subsequent recovery, to no reaction (e.g. Nielsen and Rasmussen 2009). In general, the effects of extreme events are likely greater than those resulting from any change in climate averages (e.g. Easterling et al. 2000; Jentsch et al. 2007; Zimmermann et al. 2009).
Among ecological traits of interest is phenology, which tracks annually recurring events in ecosystems such as plant germination, flowering, growth and fruit maturation. Since these events are triggered predominantly by temperature, phenology has emerged as a key asset in identifying current fingerprints of climate change in nature, especially since recent warming is mirrored by significantly advancing spring events of generally about 2–5 days decade−1 in the northern hemisphere (e.g. Menzel and Fabian 1999; Walther et al. 2002; Root et al. 2003; Matsumoto et al. 2003; Walther 2004; Menzel et al. 2006a). The climate signal controlling spring and summer phenology is fairly well understood: nearly all phenophases correlate with temperatures in the preceding 1–3 months (Sparks et al. 2000; Sparks and Menzel 2002; Menzel 2003; Cleland et al. 2007). Globally, and at continental scales, the observed changes can be attributed to human induced global warming (e.g. Root et al. 2003; Parmesan and Yohe 2003; Rosenzweig et al. 2007, 2008).
However, current phenological studies have predominantly addressed changes in “mean or average” onset dates and have analysed their variability and temporal trends in the last five decades. Only a few more recent publications have described the effects of naturally occurring or manipulated extremes on phenology (e.g. Luterbacher et al. 2007; Rutishauser et al. 2008; Jentsch et al. 2009). The study of extreme phenological events seems to be important since they may lead to critical disturbances and mismatches in ecosystems (Parmesan 2006). In particular, the response of ecosystems is most sensitive to extreme events, not average conditions, and thus events, not trends, matter (Jentsch et al. 2009; Zimmermann et al. 2009).
This study will not concentrate on whether extreme phenological events change in frequency and magnitude, but evaluates whether and to what extent recent warm and cold spells in Europe from 1951 to 2006 have resulted in extreme phenological onset dates. This analysis will also include agricultural events, so-called false phenological phases, which are driven by farm management and thus may clearly differ in their response from spring/summer events in wild nature (Menzel et al. 2006b; Estrella et al. 2007).
Materials and methods
Classification of warm and cold spells
The warm and cold events likely to coherently influence phenological responses at regional to continental scales are certainly longer than a few days, and are apt to be reflected in monthly and seasonal temperature anomalies. To classify warm and cold spells, we used quality-controlled, high-resolution gridded (0.5°) monthly and seasonal (DJF, MAM, JJA, SON) temperature means provided by ENSEMBLES (Ensembles-Based Predictions of Climate Changes and their Impacts–a project funded by the European Commission: EU FP6, http://www.ensembles-eu.org, data provided through http://eca.knmi.nl). The geographic domain was Europe west of 40°E between 36.25 and 70.75°N and the study period was 1951–2006. We used the criteria of exceeding +1.5, +3, −1.5 and −3 standard deviations from the long-term mean at the respective grid point to classify warm and cold spells into four categories of warm, very warm, cold and very cold, respectively. We defined "continental" warm or cold spells as those in which >40% of the grid points were classified as either warm or cold, and "regional" warm or cold spells if >1% were classified as either very warm or very cold.
In this study, we used the phenological data collected by the COST725 Action to assess the impacts of warm and cold spells on European plant phenology (www.cost725.org). To date, 20 European countries contributed more than 7 million phenological data values to its database covering 64 species and 22 different phases. We restricted our analysis to the time period 1951–2004 due to the relatively better spatial coverage. A constraint in the analyses was the lumped density of the almost 8,000 observational sites across countries in Central and Eastern Europe with data from several countries lacking completely (e.g. UK, Italy) or being represented only by stations from the network of the International Phenological Gardens (Menzel and Fabian 1999; Chmielewski and Rötzer 2001). In consequence, in those countries where national networks provided not only selected stations but their entire dataset, a huge concentration of stations was present, e.g. in Germany, Austria, Switzerland, Czech Republic and Estonia. In order to study the impacts of warm and cold spells on European phenology, we used onset dates of phenological seasons, such as early spring and full spring, for the large-scale comparison, since species and phases observed in the various national networks strongly differ.
Baseline of mean onsets of phenological seasons
Phenological seasons and their key phases according to Schnelle’s maps of 1965, the best corresponding phenological phases of the COST725 database and the climate triggers studied
Earliest to early spring
Middle of early summer
End of full summer
Later part of full autumn
Key phenophase (Schnelle 1965)
Sowing of summer cereals
Flowering of apple
Heading of winter wheat/cereals
Harvest of winter wheat/cereals
Sowing of winter wheat/cereals
Mean onset in Europe (DOY)
Corresponding phenological phase in the COST725 data base
Sowing/drilling of spring barley (Hordeum vulgare L.) and common oat (Avena sativa L.)
First flowers and full flowering of apple (Malus x domestica L.), both early and late cultivars
Beginning and middle of heading of winter wheat (Triticum aestivum L.) as well as of winter barley (Hordeum vulgare L.)
Harvest of winter wheat (Triticum aestivum L.), winter barley (Hordeum vulgare L.) and winter rye (Secale cereale)
Sowing/drilling of winter wheat (Triticum aestivum L.), winter barley (Hordeum vulgare L.) and winter rye (Secale cereale)
Triggering mean monthly and seasonal temperatures
In order to derive digital information from these maps, they were scanned into a rasterformat. For data processing, a highly novel procedure in Definiens Developer, accurately defined with rulesets, was set up. First, three different image object levels were created with related segmentation algorithms based on the pixel level. Within these levels, comprehensive and customised classification algorithms were used in a logical sequence to clearly separate the different classes. Finally, all background artefacts and redundant mapping, such as the grid, altitudinal shading, and annotations, were eliminated. Afterwards, the different classes were exported as a rasterformat into an ArcGIS 9.3 environment and transformed into vector data. Thus, we received digitised isoline information on phenological onset dates as shape files. Finally, the maps were geo-referenced in ArcGIS 9.3 and the onset dates were re-assembled on a 0.5° grid. Figure 1b displays the digitised GIS information of 1930–1939 mean onset dates of full spring.
The best corresponding species and phenological phases of COST725 (see Table 1) were carefully chosen as a compromise between accuracy provided by identical species and phases and increased spatial and temporal coverage granted by additional corresponding phases. Triggering months and seasons were defined according to mean onset dates in Europe.
Assessment of impacts of warm and cold spells on phenological seasons
Departures of annual phenological onsets in 1951 to 2004 from the 1930–1939 baselines were plotted against the monthly and seasonal temperature anomalies for the corresponding grid points, and expressed as confidence ellipses for bivariate normally distributed data covering 90% of data points (based on the SAS macro “ellipses” www.datavis.ca/sasmac/ellipses.html). The total number of data points, depending on season and warm/cold spells, ranged from 16 to 6,009; however, 58 out of 72 ellipses each represent more than 1,000 observations in Europe for a specific warm/cold spell. Confidence ellipses may be used as visual indicators of correlations, since they collapse diagonally when there is a high correlation between two variables, whereas they are more circular when the variables are uncorrelated. We regard the amount of variance explained by linear regressions of phenological impacts against temperature deviation (R2) as a measure of spatial coherence.
Warm and cold spells in Europe
In contrast, the (almost) two last decades of our study period (1989–2006) were mainly characterised by warm spells in all seasons, predominantly in summer and autumn, and only a few months in autumn exhibited regional cold spells. Key examples included the famous Central European heat wave of summer 2003 (up to 5°C warmer in Western and Central Europe) as well as February 1990 (up to 6°C warmer in Central Europe and around the Baltic Sea).
Analysis of the phenological impacts of warm and cold spells
Sowing of summer cereals in spring (Fig. 4a) can be regarded as a false phenological phase which mainly reflect farmers’ management decisions, and finally may or may not be reflecting what is happening climatically or biologically (see Menzel et al. 2006a, b; Estrella et al. 2007). Sowing of summer cereals, however, seemed to successfully mirror warm and cold spells of January to March and of winter. If warm spells occurred in the phenological study area of Central Europe, sowing was almost exclusively earlier by up to 30 days, and cold spells were manifested in later sowing of spring cereals by up to 20 days. Three monthly and seasonal cold spells, February 1956, January 1957 and winter 1957, demonstrated meaningful spatial coherence with observed phenological impacts (R2 of 40, 17 and 59%, respectively).
Annual anomalies of flowering dates of apple, a true phenological phase predominantly triggered by climate conditions, did mirror the warm and cold spells quite well (Fig. 4b). The cold spring of 1955, as well as cold February 1956, were linked to later apple flowering dates. Warm February and March 1990, as well as March 1989, corresponded to earlier flowering dates. Ten out of 11 warm/cold spells studied revealed a negative spatial coherence with cooler deviations being linked to later phenology. However, only for the five warm spells of 1989 (March, April, spring) and 1990 (February and March) could this spatial coherence be regarded as meaningful with larger R2 from 21 to 52%.
The picture for heading of winter cereals was unclear, mostly due to fewer identified warm and cold spells in spring months and the spring season (Fig. 4c). Only two warm spells in March 1989 and 2001 exhibited spatial coherence (R2 of 59 and 53%, respectively); however, with oppositely directed relationships.
Most interestingly, the harvest of winter cereals (Fig. 4d), which can also be regarded as a false phenological phase, also seemed to be a good indicator of climate extremes. The cold spell in summer 1962 was manifested in predominantly later harvest dates in central Europe by up to 40 days, while the summer European heat wave of 2003 as well as June 2003 were nearly exclusively linked to earlier harvest dates by up to 40 days. However, none of the warm/cold spells displayed in Fig. 4d exhibited meaningful spatial coherence between temperature and phenological anomalies.
Sowing of winter wheat (Fig. 4e) seemed to be the classical example of a false phenological phase. Regardless of a warm or cold spell inside or outside Central Europe, the sowing of winter wheat, barley and rye was up to 50 days earlier than average winter wheat sowing dates in the standard reference period (1930–1939). Only sowing dates of winter cereals in 1976 seemed to reasonably mirror the regional cold spell of October 1976 on a spatial scale (linear regression, p < 0.0001, R2 = 15%), while for all other warm/cold spells there was no spatial relationship between phenological impact and temperature deviation.
Discussion and conclusions
Based on various methods proposed in the literature for defining warm and cold spells (e.g. Beniston 2005, 2009), we defined the minimum length of a potential period in which a notable spell could occur as 1 month since shorter periods were less comparable in their effects on phenology at an European scale. Despite this coarse temporal resolution, the method presented in this paper of defining a continental or regional warm or cold spell via relative temperature deviations from a long-term reference (1) exceeding 1.5 or 3 standard deviations and (2) covering a considerable spatial extent of more than 40% of grid points or more than 1% of grid points, proved to adequately describe longer periods of extreme temperatures. The results clearly revealed that, during the non-stationary conditions in recent decades triggered by anthropogenic warming, the frequencies of warm and cold spells changed with time. Cold spells predominated until 1988, and afterwards, warm spells prevailed. Nonetheless, these two periods were not completely distinct, as in the first period regional warm spells were also noted. For the obviously warmer period of the last two decades, few regional and continental cold spells were identified.
Our analysis of the impacts of these warm and cold spells on European plant phenology relied on a mix of different species, phenophases, and datasets, ranging from point observational data to digitized maps. Can the results be regarded as robust in the face of potential pitfalls involved in using these datasets? Studying phenological seasons defined by several phenophases is a common, well-accepted procedure (Schwartz 2003). Similarly, the spatial matching of point and vector data is an accepted practice. Thus, we believe that the standard reference period of 1930–1939 based on maps drawn by Schnelle in 1965, and inclusion of other substitute phases (winter cereals) in addition to winter wheat, was not a major constraint in our study. However, the results may have suffered from a general lack of available data on the European scale and the spatial concentration of analysed sites to mainly Central and Eastern Europe. This effect can be observed in Fig. 4 where confidence ellipses without significant temperature anomaly are displayed due to the fact that phenological data were lacking in the area where a continental or regional warm/cold spell was identified.
The main topic of this paper, the question of whether and how these identified warm/cold spells translated into phenological (extreme) impacts can be assessed from two directions: firstly, studying which phenological phases and species were most affected and, in contrast, which seemed to be more resilient according to subsequent change in their onset dates, and secondly, by analysing which key phenological phases were more suitable to also mirror extreme temperature spells as a proxy for paleo-conditions. In particular for the latter case, different indicators should be checked for (1) their sensitivity, i.e. the strength of the observed advances and delays, and their consistency with the type of extreme (e.g. earlier onsets linked to warm spells), (2) the exclusiveness of the signal, and (3) the spatial coherence, i.e. whether phenological changes also spatially mimic the respective warm/cold spell.
Only a few studies on phenological extremes have so far reported species' differences in impacts of extreme warm spells. Rutishauser et al. (2008), for example, studied the effect of the very warm spring (March to May) of 2007 on 302 different phenophases and revealed that roughly one-third were especially sensitive in Switzerland. Luterbacher et al. (2007) also studied the impact of one warm spell in autumn and winter of 2006/2007. They reported phenological impacts related to this warmth such as unusual partial second flowering or extended flowering until the beginning of winter. In addition, early flowering species in spring also exhibited distinct earlier flowering after this warm winter of 2007. Our study equally reveals differences in the sensitivity of phenophases/phenological seasons to warm and cold spells. The major results revealed that the harvest of winter cereals, linked to summer temperatures, and flowering of apple, linked to early spring temperatures, were the most promising when studying the impacts of warm and cold spells in Europe. Caution should be applied if false phenological phases are used as indicators for impacts of extreme events; sometimes they appeared to be suitable, e.g. harvest of winter cereals, sowing of summer cereals, in contrast to sowing of winter cereals which was largely unaffected by any change in autumn extreme temperatures (see also Menzel et al. 2006b). Most interestingly, and only for apple flowering in spring, the warm and cold spells in February to April, as well as in the spring season also exhibited a spatial coherence with the observed phenological anomalies, i.e. the amount of extreme temperature was also mirrored in phenology.
What about differences in the sensitivity of phenological data to warm and cold spells? Using very long-term records of three tree species (300 years), Rutishauser et al. (2008) studied changes in their sensitivity, revealing distinct periods with stronger sensitivity in the cooler periods and decreased sensitivity during two periods with warming trends (1890–1950, 1970–2007). The summer heat wave of 2003 may be a promising case study for changes in sensitivity. Although phenological impacts of the European summer heat wave of 2003 have not so far been studied on a broader scale, except very recently by Garcia-Herrera et al. (2010), grape harvest dates have been used to prove the uniqueness of this heat wave in a historical context, clearly demonstrating a general sensitivity of grape harvest dates to summer temperatures (Chuine et al. 2004; Menzel 2005). However, our confidence ellipses of summer and June 2003 temperature against anomalies of harvest of winter wheat indicated a reduced response compared to other events. Further studies are needed to clarify whether a reduced sensitivity can be observed in this special year (2003); in other words, the observed harvest dates did not seem to allow a reconstruction of the exact extent of this extreme event (see also Keenan 2007).
Little attention has been paid so far to the spatial coherence of the phenological signal, i.e. whether their variations are synchronous over the same spatial domain as the temperature variations. This issue of spatial synchrony can very easily also be studied by the confidence ellipses. Diagonal ellipses, such as for the response of flowering of apple to warm spells in February/March 1990 (see Fig. 4b) or of harvest of winter wheat to the warm spell of June 2003 (Fig. 4d), indicated those cases where a spatial reconstruction of the extent and amount of the temperature extreme should also be possible.
We conclude that continental cold spells during the growing season were linked post-1989 to variable impacts on different species and phenophases. True and a few false phenological phases seem to be suitable indicators for warm and cold spells in different seasons, especially apple flowering and harvest of winter cereals; however, more research is needed to assess the value of these proxies in creating an absolute reconstruction of very extreme events.