500 years of regional forest growth variability and links to climatic extreme events in Europe

Climatic extreme events strongly affect forest growth and thus significantly influence the inter-annual terrestrial carbon balance. As we are facing an increase in frequency and intensity of climate extremes, extensive empirical archives are required to assess continental scale impacts of temperature and precipitation anomalies. Here we divide a tree-ring network of approximately 1000 sites into fifteen groups of similar high-frequency growth variability to reconstruct regional positive and negative extreme events in different parts of Europe between 1500 and 2008. Synchronized growth maxima or minima within and among regions indicate eighteen years in the pre-instrumental period and two events in the 20th century (1948, 1976) with extensive radial growth fluctuations. Comparisons with instrumental data showed that the European tree-ring network mirrors the spatial extent of temperature and precipitation extremes, but the interpretation of pre-instrumental events is challenged by lagged responses to off-growing season climate extremes. We were able to attribute growth minima in subsequent years to unfavourable August–October conditions and to mild climate during winter months associated with respiratory carbon losses. Our results emphasize the importance of carry-over effects and species-specific growth characteristics for forest productivity. Furthermore, they promote the use of regional tree-ring chronologies in research related to climate variability and terrestrial carbon sink dynamics.


Introduction
. Lastly, carry-over effects from the previous growing season significantly influence growth performance (Fritts 1976, Frank et al 2007a and may thus mitigate or enhance impacts of seasonal climate extremes on forest productivity-an issue not considered in most ecophysiological investigations. Extensive and precisely dated empirical datasets are required to quantify large-scale forest ecosystem performance before, during and after extreme climate events. Measurements of the annual radial growth increment (tree-rings) offer such possibilities by allowing forest growth to be related to climate variability at local (Panayotov et al 2011), regional (Neuwirth et al 2007) and continental (Briffa et al 2002) scales. Furthermore, tree-ring chronologies can preserve growth variations over decades to millennia (Büntgen et al 2011) and thus can record adequate numbers of extreme events, which are rare by definition (Battipaglia et al 2010). Yet, effects from climate extremes occurring outside the main growing season, which may indirectly influence radial growth via nutrient storage (Kreuzwieser and Gessler 2010), altered phenological phases or changes in growing season length (Moser et al 2010), are usually more difficult to detect in the respective annual ring. Such climatic events may be recorded as carry-over effects leading to a strong growth reaction in the following year (Fritts 1976), however, the exact circumstances leading to lagged growth extremes have not yet been fully clarified on a large scale (Wettstein et al 2011). Improved understanding of both the immediate and lagged consequences of extreme events are crucial for interpreting growth anomalies during the pre-instrumental period as well as reducing uncertainties in ecosystem projections.
Here, we employ a continental scale tree-ring network (Babst et al 2012) to assess and reconstruct years with extremely high or low forest growth in different parts of Europe. We divide the network into sectors of similar annual variability and calculate mean tree-ring chronologies reflecting forest growth anomalies over the past five centuries. Furthermore, we compare extreme growth years at approximately 1000 sites to temperature and precipitation anomalies across Europe and test, if tree-ring networks reflect the spatial extent of climate extremes. Thereby we address extreme growth between 1500 and 2008 at sites predominantly sensitive to temperature, precipitation and mixed climatic influences. Additionally, we employ monthly temperature and precipitation data in extreme years to differentiate climate conditions leading to contemporaneous growth extremes from those conditions causing anomalous growth also or only in the subsequent year.

Tree-ring and climate data
We used a large tree-ring width (TRW) network of approximately 1000 sites (Babst et al 2012) of the most abundant species to assess annual forest growth variability across Europe. For the assessed time periods, all TRW chronologies had a minimum sample replication of five series and an expressed population signal (EPS) above 0.85 (Wigley et al 1984) making the network well replicated over the past five centuries. To isolate extreme growth years in tree-ring data, it is necessary to remove low-frequency signals, including the biological age trend (Fritts 1976) and long-term growth changes (Esper et al 2002). This was accomplished using a spline detrending with a 50% frequency cutoff response at 30 years which preserves inter-annual to decadal variability in the resulting standard chronologies.
For climatic analyses, we derived monthly temperature and precipitation data from a fine-spatial (1×1 km) resolution gridded climatology spanning the 1901-2006 period (see Babst et al 2012 for a description). These instrumental data were detrended analogously to the tree-ring data to extract higher frequency variability.

Regional tree-ring width chronologies
We produced regionally representative tree-ring chronologies as the basis to assess and reconstruct growth extremes in different parts of Europe. Accordingly, we divided the European TRW network into fifteen groups of similar annual growth variability using a fuzzy clustering algorithm (Kaufman and Rousseeuw 1990), thereby disregarding any spatial information. The number of clusters was determined based upon sensitivity analyses of mean chronology intercorrelations (rbar) as well as cluster replication/spatial representation (appendix A). In contrast to a 'hard' k-means clustering which produces identical groups, the fuzzy approach additionally provides the degree of membership of a site to each of the fifteen clusters. Thus, when calculating representative mean curves, we were able to weight the influence of each site on the resulting regional tree-ring chronologies (RTCs) by primary cluster membership to best represent the integrated forest area. Additionally, we accounted for changes in sample replication and rbar to minimize possible variance artefacts related to changes in the network composition (Frank et al 2007b) and truncated RTCs where the replication dropped below five sites (figure 1). We assessed monthly climate sensitivities of each RTC between April of the year prior to ring formation and current September (appendix B). Furthermore, negative or positive growth extremes were defined when TRW was below the 5th or exceeded the 95th percentile over the cluster calibration (1920-70) and verification (1870-1919) periods (Stefanon et al 2012).

Spatial climate anomalies and lagged growth extremes
We compared growth extremes indicated by the tree-ring network to seasonal climate anomalies across Europe to assess common spatial patterns. This comparison was done at the site level since climate anomalies do not necessarily affect the entire areas represented by an RTC, which may therefore imprecisely reflect the spatial extent of e.g. heat waves. Each site of the tree-ring network was classified according to significant (p < 0.05) correlations with instrumental data as temperature (r temp > 0.273, 232 sites), precipitation (r precip > 0.273, 216 sites), or mixed (0.273 > r temp, precip > −0.273, 544 sites) sensitive. The mixed category contains sites which are neither strongly temperature nor water limited during the April-July period which was identified as the primary season influencing forest growth integrated over Europe (appendix B). Lagged growth extremes in the following year, however, cannot be explained by April-July climate anomalies only. Thus, we employed monthly temperature and precipitation indices in conjunction with a simple decision tree model (figure C.1) to distinguish climate conditions leading to immediate growth extremes from those conditions causing lag effects. Subsequently we use the following abbreviations to distinguish different types of extreme responses: C-EX = growth extreme in the current year; CL-EX = growth extreme in the current and in the following year; L-EX = growth extreme in the following year.

Regional forest growth variability
We produced fifteen RTCs representing high-frequency growth variability of forests across Europe over the past 200-800 years. These contain between 41 and 103 site chronologies with mean inter-site correlations usually above 0.3 (see table D.1). Each RTC integrates a confined latitudinal, longitudinal and altitudinal space which emerged from a clustering solely based on annual radial growth variability (figure 2). While the represented areas overlap considerably, the majority of RTCs are dominated by no more than two species (table 1). The separation of RTCs primarily emerges from differing growth patterns under normal climatic conditions. RTCs may, however, respond synchronously to climate extremes within and among regions indicating years with extensive productivity maxima or minima.  For better visibility, the high-elevation Alpine RTCs 10-12 are shown in a separate inset.

Extreme growth events between 1500 and 2008
We reconstructed positive and negative growth extremes across Europe over the past 500 years based on RTCs (figure 3  (1948 and 1976) where forest growth was extremely high or low over large areas in Europe (figure 3 bottom).

Climate and growth anomalies in space and time
We assessed the spatial extent of forest growth extremes between 1500 and 2008 by mapping their annual occurrence and sign together with the expected climate sensitivity. Maps for each year during the full time period are provided in the supplementary material (available at stacks.iop.org/ERL/7/ 045705/mmedia). Direct comparisons with April-July climate anomalies were performed over the instrumental period  with a focus on 1976 and 1948 (figure 4)-the most extreme years in the RTCs. In 1976, exceptionally hot and dry conditions exceeding two standard deviations prevailed over central and western Europe. Furthermore, warm and wet conditions occurred on the Iberian peninsula whereas in Scandinavia and eastern Europe precipitation amounts were low. The growth impacts of this climate pattern largely agrees with the signals expected from the site specific climate responses. In central and eastern Europe, precipitation and mixed-sensitivity sites showed a negative growth extreme in response to intense drought, while mid-elevation mixed-sensitivity sites on the Iberian Peninsula benefited from warm and pluvial conditions. Most temperature sensitive sites in the Alps showed no strong growth increase during the 1976 heat episode, likely due to legacies from low growth in 1974 (see small plot in figure 4) and 1975 (see supplementary material, available at stacks.iop. org/ERL/7/045705/mmedia). In 1948, low growth occurred at many precipitation and mixed-sensitivity sites in central Europe and the low-elevation Alps, suggesting drought. This signal, however, appears to result from lagged effects driven by the previous summer drought. In 1947, low growth of forests in southern, southeastern and central Europe reflected the dry conditions whereas high-elevation sites in the Alps benefited from warm temperatures (see small plot in figure 4 and supplementary material, available at stacks.iop.org/ERL/ 7/045705/mmedia). Positive growth was also observed in Great Britain due to enhanced moisture availability as well as in Scandinavia likely attributable to the expansive warmth in this year.
In the pre-instrumental period, spatial patterns of reconstructed growth extremes become less distinct with decreasing site replication back in time. Yet-while we refrain from making direct comparisons with early instrumental data (e.g. Böhm et al 2010)-it is clear from our results that climatically driven and spatially pervasive extensive growth extremes occurred in Europe during the past centuries.

Contemporaneous and lagged growth extremes
We employed the European tree-ring network and monthly instrumental data in an attempt to distinguish between climate conditions leading to contemporaneous growth extremes (C-EX) and those causing lag effects (CL-EX, L-EX). A separate assessment at temperature, precipitation and mixed sensitive sites, however, did not produce robust results due to the rare occurrence of CL-EX situations during the instrumental period. Thus, we combined temperature and precipitation limited sites where we compiled monthly indices of the respective limiting climate factor in all extreme growth years. Thereby we distinguished between C-EX, CL-EX and L-EX events and calculated bootstrapped means (Boos 2003) of the monthly climate indices leading to each of these three growth extreme types. Figure 5 shows the 95% confidence intervals of the means of monthly climate indices causing positive and negative growth extremes, as well as normal growth.
The robust patterns in the seasonal course of temperature and precipitation variations revealed significant differences in climate causing isolated and lagged growth extremes. C-EX growth responses (both positive and negative) are generally triggered by anomalous climate during the April-July season. In particular, ideal conditions in July lead to enhanced growth, while productivity drops depend upon a broader season. Negative CL-EX and L-EX events on the other hand are related to August-October climate. This relationship indicates, that unfavourable late summer to early autumn conditions leads to lower forest productivity in the following year. Additionally, reduced growth limitations in winter (December-February) contribute to low growth in the subsequent summer if previous growing season climate was moderately (negative L-EX) or strongly (negative CL-EX) disadvantageous. Positive L-EX events are caused by two consecutive growing seasons with beneficial climate conditions while no clear pattern leading to positive CL-EX events was found.

Discussion
Understanding the impacts of climate extremes on European forests requires information on tree growth at large scales. The RTCs developed in this study integrate high-frequency  figure 3)-are dominated by individual tree species emphasizing species-specific growth characteristics under comparable climatic conditions. Thus, growth extremes of underrepresented tree species, and the extent to which mixed forests may be more resilient towards climatic extreme events, are not fully assessable using the current network. RTCs represent surprisingly confined areas, even though no spatial information was considered in the underlying grouping. Thus, forest productivity in the respective areas is expected to follow similar climatic drivers. Over long periods, however, the climate sensitivity of forests may shift due to low-frequency trends such as the 20th century warming (Lloyd and Bunn 2007, Andreu et al 2007) or changes in the atmospheric composition (Phillips et al 2008). Even though the magnitude of such variations in external forcing likely varies on a continental scale, we observed synchronous growth extremes between 1500 and 2008 within and among regions. This suggests that annual growth changes during the calibration and verification periods of RTCs were reasonably consistent with inter-annual to multi-decadal growth variability over five centuries. We identified extremely high and low growth across Europe between 1500 and 2008 (see supplementary material available at stacks.iop.org/ERL/7/045705/mmedia). Largescale growth extremes are rare due to the offsetting responses of temperature, precipitation and mixed-sensitivity sites as well as the physical modes of the climate system with contrasting regimes during summer in southern versus Comparisons with climate anomalies during the 20th century suggest, that large tree-ring compilations mirror the spatial extent of temperature and precipitation extremes. A direct interpretation of pre-instrumental growth extremes is, however, complicated by seasonal variations in forests' climate responses (Babst et al 2012) and carry-over effects from the previous year (this study, Frank et al 2007a). Many regional dendroclimatological studies found positive correlations between radial tree growth and previous summer to autumn climate (e.g. Carrer et al 2010) and we observed a similar climate response of the fifteen RTCs (appendix B). It is important to note, however, that the species represented in our tree-ring network follow determinate growth mechanisms and therefore are significantly influenced by environmental conditions during bud formation (Borchert 1991). Carry-over effects from the previous growing season are likely enhanced compared to species which exert indeterminate growth, but testing this hypothesis would require a vastly enlarged network. For the major European tree species, our results emphasize the importance of lagged growth extremes, as evidenced by the growth anomalies during the 1947/48 and 1976 heat and drought events in central Europe and the high-elevation Alps respectively. Importantly, we were able to reveal how characteristic patterns in temperature and precipitation anomalies over the course of more than 18 months cause immediate or lagged growth reactions.
In our continental scale tree-ring network, we attributed growth minima in the subsequent year to negative August-October extremes of limiting climate parameters at temperature and precipitation responsive sites. This result emphasizes the influence of autumn climate on forest growth-a factor which is expected to become even more important due to changes in ecosystem respiration caused by warming trends (Vesala et al 2010). Furthermore, productivity in the subsequent year appears to be lower if growth limitations are reduced during the cold season. This signal presumably derives from enhanced respiratory carbon losses at temperature limited sites caused by warm conditions outside the main growing season (Piao et al 2008). These findings contribute to the mechanistic understanding of ring formation and also are relevant for terrestrial carbon sink estimates based on dynamic global vegetation models which currently do not sufficiently consider carry-over effects in simulations of ecosystem productivity (Sitch et al 2008).

Conclusions
High-frequency variability of European forest growth over the past 500 years is expressed by fifteen robust RTCs which are mainly separated by geographic position, climate response and species-specific growth characteristics. RTCs respond synchronously to extensive climate anomalies and indicate extreme years which are of particular interest for comparison with other data streams (e.g. crop yield). The spatial extent of strong climate anomalies is reflected by a continental scale tree-ring network, however, the direct attribution of past growth extremes to specific climatic conditions remains challenged by the intermittent occurrence of lagged responses to climate anomalies outside the current growing season. We could determine that negative lag effects are mainly caused by unfavourable August-October climatic conditions at temperature and precipitation limited sites. Our results demonstrate the merits of large empirical networks to assess the impacts of climate extremes on forest growth. In particular we provided the first continental-wide reconstruction of the timing and spatial manifestations of growth extremes during the past 500 years where the inter-annual terrestrial carbon balance was likely significantly affected. These are the first efforts to provide a long-term benchmark to place impacts   Correlations between fifteen regional tree-ring chronologies (RTCs) and monthly temperature and precipitation data. Correlations are shown between April of the year prior to ring formation and current September. Black rectangles highlight significant months (r > 0.273, 95% confidence level). RTCs are grouped according to the geographic space covered (NE: northern Europe, CE: central Europe, AL: Alps low elevation <1300 m a.s.l., AH: Alps high elevation >1300 m a.s.l., SE: southern Europe). precipitation data analogue to the calculation of RTCs. From monthly climate responses of clusters, we identified seasons which strongly influence forest growth in different parts of Europe. Additionally, we performed a linear regression analysis between the average previous April to current September climate correlations and mean annual temperature and precipitation of each cluster to identify the strongest common drivers of forest growth across the continent.
Results show that climate during the April-July season commonly influences forest growth across Europe. On a regional scale, however, the seasonality in the climate response varies strongly (figure B.1). July temperature is the strongest driver of forest productivity in northern Scandinavia (RTC 1) whereas in southern Scandinavia (RTC 2), spring to summer precipitation and spring temperature positively influence wood formation. Additionally, high temperatures during previous July-August hamper growth in this area. Forests in central Europe (RTCs 3-5) as well as at low to mid-elevation Alpine regions (RTCs 6-9) respond positively to April-July and in the case of RTCs 4 and 6 also to previous