Long term January – March and May – August temperature reconstructions from tree-ring records from Bosnia and Herzegovina

Introduction Conclusions References

western part of the Balkan Peninsula as a mixture of continental climatic influence from the interior of the peninsula, mountain climatic influence from the Dinaric Alps and Mediterranean influence from the Adriatic Sea. Annual precipitation and temperature regimes are characterized by seasonally diverse circulation patterns. In spring, an Atlantic High extending eastwards and over the Balkan Peninsula joins a low 10 centre approaching from the southeast, causing a north-northeasterly flow over the Eastern Mediterranean area. The extension of the summer Asian thermal low is evident throughout the Eastern Mediterranean in all summer circulation patterns; however, it controls the weather in the region jointly with other principal pressure features (Kostopoulou and Jones, 2007). 15 In the light of expected climate change, heat waves in the Mediterranean region and on the Balkan Peninsula will intensify in the second half of the 21st century -they will be more frequent and will last longer. According to predictions, the minimum daily temperature during the worst heat events is expected to rise by around 3 • C ( Meehl and Tebaldi, 2004). In the Eastern Mediterranean and the Middle East, there will be a grad- 20 ual and relatively strong warming of about 3.5-7 • C between the 1961-1990 reference period and the period 2070-2099 (Lelieveld et al., 2012). The observed daytime maximum temperatures appear to be increasing most rapidly in the northern part of the region, i.e. the Balkan Peninsula and Turkey. Hot summer conditions that rarely occurred in the reference period may become the norm by the middle or end of the 21st 25 century (Lelieveld et al., 2012). Moreover, a decrease in annual mean precipitation from −10 % to more than −20 % over some regions of the Mediterranean basin is expected by the end of the 21st century (IPCC, 2001 basin the observed strong drought period of the end of the twentieth century seems to be the strongest of the last 500 yr (Nicault et al., 2008). It is therefore important to investigate whether extreme events have already occurred in the past and how trees responded to them. 5 Selecting tree species growing on sites with limited between-tree competition and climate as the prevailing growth-limiting factor maximises the climate signal in tree-rings. P. nigra was chosen for this study, since it grows on extreme sites, has a good response to climate and reaches ages up to 500 years (Brus, 2004). The area of distribution of P. nigra covers the majority of the Mediterranean region (Vidaković, 1991), so the results of the climate-growth relationship from BiH can be compared to results from other regions. Its growth response to climate has been used in many climate-growth studies in the western (Martín-Benito et al., 2011, 2010b and Eastern Mediterranean (Sevgi and Akkemik, 2007;Touchan et al., 2003), in the northern limit of its natural arealin Austria (Leal et al., 2008) and North-Eastern Romania (Levanič, 2012). Since the 15 species is well-adapted to the Mediterranean and Southern European climate, P. nigra tolerates summer droughts and high temperatures (Penuelas and Pilella, 2003), but poorly tolerates drought during early spring (Wimmer et al., 2000).

Species selection and response to climate
In this paper, we present a climate reconstruction based on a P. nigra tree-ring width chronology from the north-western part of the Balkan Peninsula. The main goal is Introduction amount of precipitation. Despite the large amount of precipitation, the studied sites are generally dry because of surface run-off, mostly southern exposition and steep slopes. The Krivaja and Konjuh sites are close to one another, but differ in altitude and aspect (Table 1). 15 Old, healthy and dominant or co-dominant trees were selected for sampling. From each tree, two cores from opposite sides were taken at breast height (1.3 m) and perpendicular to the slope to avoid compression wood. Cores were air-dried, glued onto wooden holders and sanded with progressively finer sandpaper until a highly polished surface was achieved (Stokes and Smiley, 1996). Samples were scanned us-Introduction  (Eckstein and Bauch, 1969). Additional quality control was applied using the COFECHA program (Holmes, 1983). COFECHA assesses the quality of cross-dating and measurement accuracy of TRW series (Grissino-Mayer, 2001). If COFECHA highlighted any potential problems, the TRW were visually crosschecked and, if necessary, measurements repeated, re-checked or removed from further pro-5 cessing if recognized as unusable. The reasons for removing certain trees from the sample pool were growth anomalies connected with resin collection in the past, overgrown wounds or the occurrence of compression wood deep inside the trunk. Individual TRW series were fitted with a cubic smoothing spline with a 50 % frequency response at 67 % of the series length to remove non-climatic trends due to 10 the tree's age, size, and the effects of stand dynamics (Cook and Briffa, 1990). Standardisation was done using ARSTAN for Windows, version 4.1d, the program provided by Cook and Krusic, Lamonth-Doherty Earth Observatory, Columbia University (http://www.ldeo.columbia.edu/trl). Each year's ring width was divided by the year's value of the fitted curve to give a dimensionless index with a mean of one. Index val-15 ues were then pre-whitened using an autoregressive model selected on the basis of the minimum Akaike criterion and combined across all series using a bi-weight robust estimation of the mean to exclude the influence of outliers. Two chronologies for each site were produced in this way; a standard chronology and a residual chronology containing only high-frequency variations with statistically removed autocorrelation (Cook, 20 1985;Cook et al., 1990). The signal strength in the residual chronology was tested using the Expressed Population Signal -EPS (Wigley et al., 1984). The calculation of EPS was based on a 50-yr running window, with a 49-yr overlap. The usable portion of a chronology was defined as the part in which a minimum number of trees maintained an EPS value above 0.85 (Briffa and Jones, 1990 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | mean sensitivity (MS). An MS around 2.0 is generally accepted as indicating a series that is sensitive enough for climate reconstruction (Speer, 2010).

Climate data
We tested two different climate datasets for BiH. The first was the CRU TS3.  , Tjentište (1964Tjentište ( -1992 and Čemerno (1958-1992). We compared CRU TS3.1 to individual weather stations and found that they all show similar variations in temperature and precipitation data. In order to avoid problems with missing data in the local meteorological datasets, we decided to use a complete, 20 homogenised, crosschecked and verified CRU TS3.1 dataset for further analysis. The KNMI Climate Explorer web page enables extraction of climate data for a certain point or an area, defined by its longitude and latitude. In our case, we set the limits for BiH to E16 • 33 -E18

Statistical analysis
Pearson's correlation coefficient was used to evaluate the relationship between climate and annual radial growth of P. nigra. Residual site chronologies were compared to mean monthly temperature and the monthly sum of precipitation, to detect potential differences in a tree's response to climate at different sites. In addition to simple monthly 5 climate values for tree-ring indices comparison, we also generated a number of seasonal temperature and precipitation (not shown here) variables (e.g., May-August or January-March mean temperature), and correlated them with the tree-ring indices. The significance and confidence intervals of Pearson's correlation coefficient and the temporal stability of the relationship between climate and annual radial tree-growth were 10 calculated using a bootstrap procedure (Guiot, 1991) in R libraries dplR (Bunn, 2010) and BootRes (Zang, 2011). A linear model was used to compute the relationship between tree-ring indices and climate data. To assess the quality of the model for climate reconstruction, the period of the measured meteorological data was split into two equal periods for calibration 15 and verification (Fritts, 1976). The procedure was then repeated with the periods reversed. The reliability and prediction skill of the model was tested using mean squared error (MSE), reduction of error (RE) (Fritts, 1976), coefficient of efficiency (CE) (Cook et al., 1999) and Pearson's correlation coefficient (r). Values were computed for both calibration and verification periods. If RE and CE coefficients are higher than zero, the 20 relationship has a predictive value, and a transfer function can be calculated and applied on the regional chronology. Application of the transfer function on the chronology over a period of time for which there is no climate data results in reconstruction of the climate.
Years of extreme cold/warm springs and cool/hot summers were defined as years 25 in which the reconstructed mean January-March or May-August temperatures were below or above the specified threshold. This approach was used by Touchan (2008) in the analysis of reconstructed precipitation in Tunisia. Thresholds can be specified Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | by the final user of the reconstruction. In our case, we used 15th and 85th percentile thresholds.

Climate signal analysis
One local chronology was constructed for each sampled site in BiH. For more infor-5 mation about the chronologies, see Poljanšek et al. (2012). The temperature and precipitation (not shown here) signals in site chronologies were tested using Pearson's correlation coefficient between tree-ring indices and the CRU TS3.1 climate dataset for the period 1901-2009 from January to October (Jones and Harris, 2008). Two distinct and significant responses between site tree-ring indices and average monthly temperatures were found ( Fig. 1).
On the sites Blace, Peručica, Šator and Konjuh, we found a significant response to mean monthly January, February and March temperatures. On the other three sites, Krivaja, Prusac and Šipovo, we identified a clear response of TRW to mean monthly May, June, July and August temperatures and summed precipitation. Based on cor-15 relation analysis ( Fig. 1), we combined the site chronologies into two parts, reflecting spring and summer responses to climate. The sampled trees from the sites Blace, Peručica, Šator and Konjuh aggregated into a "SPRING" group and trees from the sites Krivaja, Prusac and Šipovo aggregated into a "SUMMER" group. Two newly developed group chronologies were developed in this way (Fig. 2). The chronology for the 20 SUMMER group, constructed from a combination of trees from the associated sites, is longer than the SPRING chronology. According to EPS values, however, the time-span of the chronology available for reconstruction is longer for the SPRING group (Table 2 In the next step, group residual chronologies were compared to monthly climate data and created aggregates of January-February-March (JFM) and May-June-July-August (MJJA) climate variables (Fig. 3). While the SPRING group chronology has a significant January-February-March signal and weak summer signal, the SUMMER group chronology has a distinct summer (May-June-July-August) and weak early year (early spring) signal. Pearson's correlation coefficients between the TRW indices and temperature were 0.40 (p < 1 %; n = 109) for JFM in the SPRING group and −0.58 (p < 1 %; n = 109) for MJJA in the SUMMER group. In the SPRING group, MJJA temperature and precipitation influence are also significant, but not stronger than in the SUMMER group. 10 The spatial strength of the SPRING and SUMMER chronologies, tested using spatial correlation in KNMI Climate Explorer (van Oldenborgh, 1999), showed that the SPRING residual chronology correlates best with the aggregated JFM average temperature (0.4 > r > 0.5, p < 0.05) in the area extending north to Hungary, east to the Black Sea, south to Greece and west to the Adriatic Sea (Fig. 4, left). The SUMMER residual 15 chronology correlates best with the aggregated MJJA temperature (−0.6 < r < −0.5, p < 0.05), covering the complete western and major central part of the Balkan Peninsula, from Greece to Hungary (Fig. 4, right). This makes our chronologies valuable proxies for large-scale climate reconstruction. 20 The correlation between TRW indices and MJJA temperatures for the SUMMER group chronology was −0.58 and with MJJA precipitation 0.55 (in both cases n = 108, p < 0.001). For the SPRING group chronology, the correlation between TRW indices and JFM temperatures was 0.40 (n = 108, p < 0.001). In addition to the statistical significance of the correlation coefficients for the whole period of available meteorological 25 data (1901-2009), we also tested the temporal stability of the relationship between temperature and precipitation data and tree-ring indices using a 31-yr running correlation, lagged by 1 yr, as previously used by Brázdil et al. (2010) and precipitation signals in the SUMMER group are stable between 1901 and 1965 (Fig. 5). After that period they start to approach a 95 % significance line and reach the level of significance between 1981 and 1982. The 31-yr running correlation revealed that the JFM temperature signal in the SPRING group is stable in the period between 1901 and 1941, when the latter repeatedly oscillates around the 95 % significance line 5 (not shown here). This indicates that the late winter-early spring signal in the TRW is weaker than the summer signal.

Transfer function development
Based on correlation analysis, we developed three linear models between the newly developed TRW indices and mean JFM, MJJA temperatures and summed MJJA pre- where Y is the mean MJJA temperature ( • C) and X represents TRW indices of the SUMMER group. The predictor variable accounts for 34 % of the variability of the climate. The second regression model, between the SUMMER group TRW indices (predictor) and the sum of MJJA precipitation (prediction), is also significant (F = 43.7, 15 p < 0.0001). The model equation is: Y = −12.76 + 347X , where Y is the sum of MJJA precipitation (mm) and X represents the TRW indices. The predictor variable accounts for 29 % of the variance. The third regression model is between the TRW indices (predictor) of the SPRING group and mean JFM temperature (prediction). The model equation is: Y = −4.74 + 4.58X , where Y is the mean JFM temperature ( • C) and X repre-20 sents the TRW indices. The model is significant (F = 17.8, p < 0.0001). The predictor variable accounts for only 14 % of the variance, which is less than with the first two models. The prediction skill and stability of all three models was verified using a splitsample procedure, by dividing the period into two subsets of equal length for calibration and verification; 1901-1955 and 1956-2009 (see Table 3 and Fig. 6).
Comparing the results of correlation analysis and the split-sample procedure between MJJA temperature and precipitation for the SUMMER group, we identified mean MJJA temperature as the most significant seasonal predictor for the reconstruction, 4412 Introduction  (Table 3). The split-sample calibration-validation exercise indicated stability of the relationship over the two halves of the available instrumental data period (Table 3 and Fig. 6). The similarity and strength of the derived calibration equations and verification tests of the two subset periods justify using the full period for developing the mean MJJA temperature reconstruction.

5
For the SPRING group, the correlation analysis and split-sample calibration-validation exercise confirm JFM temperature as the significant seasonal predictor for the reconstruction (Table 3). Since all RE and CE values are greater than 0, the transfer functions can be used for the reconstruction of JFM and MJJA temperatures. 10 The 15th and 85th percentiles of observed mean MJJA temperature ( respond to different thresholds for observed and predicted MJJA temperatures, which is a consequence of compression of variance in the reconstruction process (Touchan et al., 2008). The correlation coefficient between the reconstructed and observed temperatures is 0.58 (n = 109, p < 0.001). With a threshold set at the 15th and 85th percentiles, we were able to reconstruct 6 out of 17 observed hottest summers and 6 out 20 of 17 observed coolest summers in the available meteorological data set. The years 1945The years , 1947The years , 1950The years , 1988The years , 2000The years and 2003The years were identified as hot and 1913The years , 1926The years , 1959The years , 1960The years , 1978The years and 1984 as cool summers (Fig. 7). Reconstructed cool and hot summer periods were identified in a two-step procedure; first with a threshold for observed temperatures and then with threshold for predicted 25 temperatures. The designations of hot and cool summers discussed in subsequent sections are based on the 14.75

Identification of hot and cool summer periods
• C and 16.58 • C thresholds defined from the percentiles of the measured temperature (Fig. 7) 15 cool (1710, 1711, 1729, 1735, 1831, 1833, 1870, 1871, 1885, 1896, 1898, 1913, 1959, 1966, 1984) and 16 hot summers (1715,1725,1782,1802,1823,1835,1872,1880,1907,1945,1947,1950,1958,1968,1971,2000) (Fig. 8). According to the chosen percentiles, the springs of 1902 and 1989 were identified as warm and 1905, 1917, 1929, 1940 and 1963 as years with a cold spring (not shown here). With a threshold set at the 15th and 85th percentiles, we were able to reconstruct 2 out of 17 observed warm springs and 5 out of 16 observed cold springs in the 15 chosen meteorological data set. With 1.34 • C and −1.97 • C as thresholds, the reconstruction from the SPRING group for the period 1685-2009 contains three cold (1725,1763,1782) and three warm springs (1772, 1843, 1876) (Fig. 9). 20 Temperature reconstruction for the north-western part of the Balkan Peninsula is based on the P. nigra TRW chronologies for the BiH area (Poljanšek et al., 2012). Each site chronology was compared to CRU TS3.1 climate data for the BiH area. Correlation analysis (Fig. 1)  between Mediterranean and continental climate, similar as Pirin Mountains in southwest Bulgaria spatially mark a transition between the Mediterranean and temperate climate zones (Grunewald et al., 2009). Four out of seven site chronologies showed a significant response to spring temperatures, while the remainder showed summer temperature and precipitation signals. Based on a correlation matrix (Fig. 1), we di-5 vided the studied sites into two groups -SUMMER (Krivaja, Prusac, Šipovo) and SPRING (Blace, Peručica, Šator and Konjuh) group. The SPRING sites are located on the Mediterranean exposed wet side of the Dinaric massif and are more influenced by spring temperature, while the SUMMER sites are located in the eastern, drier side of the Dinaric massif (or in the precipitation shadow) and are more under the influence of summer temperatures (Fig. 4). The exception is the Konjuh site (in the SPRING group), since it is located in the interior of BiH but contains a significant spring temperature signal. This may be connected to the serpentine bedrock of the site, since serpentine is a water bearing rock. Krivaja is also on serpentine but its elevation is only 500 m, which may be why it has a summer climate signal. 15 In general, the annual radial growth of P. nigra in BiH is influenced positively by January to March temperatures of the current year and negatively by mean May to August temperatures, while precipitation in the period May-August has a positive affect (Fig. 3). This result is similar to that from sub-Mediterranean Slovenia, where a positive influence of winter-spring temperatures and negative influence of summer tempera-20 tures has been reported (Ogrin, 2005). January and March temperatures of the current year have no direct influence on cambium activity, but their positive influence on radial growth can be explained through water availability at the beginning of the growing period, caused by mild and wet winters. Mild  circulation and enhanced precipitation and therefore sufficient soil recharge. This has a positive influence on a wider radial increment of P. nigra. During cold winters, storm tracks from the Atlantic move towards Northern Europe, leading to dry winters in the Mediterranean, including the Western Balkans. Lopez Moreno and Vicente Serrano (2008) report that, during negative phases, the standardized precipitation index is dom-5 inantly positive in Southern Europe, including the Balkans, indicating wet conditions. Radial growth of P. nigra in the mountains of the western part of the Balkan Peninsula is negatively affected by above-average May-August temperatures and positively by precipitation in the same period. This contradicts findings from the Alps, where tree growth is mainly influenced by temperature and the influence of above-average tem-10 peratures on tree-ring formation is positive (Frank and Esper, 2005;Rossi et al., 2007;Levanič et al., 2009;Carrer et al., 2007) or in the drier Mediterranean region, where precipitation governs tree-growth (Touchan et al., 2005b(Touchan et al., , 2008. The reason why our findings are not identical to observations in other parts of the Mediterranean region, where precipitation plays an important role in tree growth and the dynamics of TRW 15 formation (Touchan et al., 2005a(Touchan et al., , 2007, may be orographic precipitation. The weaker precipitation signal in BiH comparing to those from the Mediterranean region (Martín-Benito et al., 2010b;Touchan et al., 2005b) may be a consequence of high amounts of precipitation along the Adriatic coast. Although significant, the precipitation signal in the TRW is not as strong as the temperature signal, because the Dinaric Mountains, 20 stretching along the Adriatic coast, are well supplied with water. Towards the interior of the Balkan Peninsula and in Central Europe, an increasing influence of precipitation on growth has been observed in the Vienna basin region (Strumia et al., 1997) and Romania (Levanič, 2012), where P. nigra reacts more strongly to July rainfall than to temperatures. Similar results have been reported from Turkey, where P. nigra responds 25 positively to summer precipitation, but not to temperature (Sevgi and Akkemik, 2007).

General response to climate
In contrast to the relatively uniform response to a single climatic factor, a combined response of P. nigra to precipitation and temperature has been reported from Albania and Spain, which is similar to our finding in BiH. In Albania, a significant negative Introduction response to June, July and August temperatures and positive response to June precipitation on tree radial growth has been observed, while the influence of May temperatures was just on the limit of significance (Levanič and Toromani, 2010). In Spain, July and August temperature and precipitation have a significant effect on tree-ring formation (Martín-Benito et al., 2010a). Our findings are therefore more in accordance with 5 results from Albania and Spain than those from other countries. The SUMMER group contains stable and significant temperature (r = −0.58) and precipitation (r = 0.40) signals (Fig. 3). Signals are stable from the beginning of the climate CRU dataset, until the end of the 1960s (Fig. 5). Weakening of the signal occurs simultaneously with changes in summer precipitation and temperature decadal 10 variations. The period of decreased sensitivity of trees to summer temperature and precipitation starts at the end of the 1960s and lasts to the beginning of the 1990s. According to Mariotti and Dell'Aquila (2012), this period was characterized by outstanding decadal variations of summer temperature over the entire Mediterranean region. The coolest periods in the Mediterranean region after 1950 were during the 1960s and 1970s, with the coolest summer in 1976 and the warmest summers in the 1990s. Mediterranean summer temperature anomalies were also very well reflected in BiH, where the coolest summer was recorded in 1974 and the warmest summers in 1994, 1998 and 2000. The cooling trend for the period 1950 to 1960 is −1.15 whereas a trend of 0.5 • C decade −1 was recorded for the period between 1980 and 20 1999 (Xoplaki et al., 2003). There is also a clear tendency for wetter summers between 1967 and 1985 over the Balkan region, when compared to previous decades (Blade et al., 2011), while the period after 1985 was characterized by drier summers. It should be emphasized, though, that the drying trend after 1970 was less pronounced than the wetting trend two decades earlier. 25 Precipitation variations on decadal and multi-decadal scales are significantly affected by the variability of the North Atlantic Oscillation (NAO), which modifies sea level pressure and associated circulation anomalies (Mariotti and Dell'Aquila, 2012). Summer temperature, on the other hand, is significantly influenced by the Atlantic multi-decadal CPD 8,2012 Spring and summer temperature reconstruction for the NW Balkan Oscillation (a mode of variability occurring in the North Atlantic Ocean and which has its principal expression in the sea surface temperature field). Summer NAO (SNAO) explains around 30 % of decadal precipitation variability over the Balkans (Folland et al., 2009). It has been shown that the existence of upper level troughs over the Balkans, Italy and Spain can explain the SNAO influence on summer precipitation (Blade et al., More local patterns indicate that the number of summer cyclones in the central part of the Adriatic Sea increased significantly after 1970, affecting precipitation in the Western Balkans (Bartholy et al., 2009). This concept is important, since large scale patterns are important in triggering local summer convection. Regional processes and feed- 15 backs modulate the influence of large scale anomalies over the Mediterranean during summer. Mariotti and Dell'Aquilla (2012) have shown that these processes may involve cloud cover, land surface modifications and include positive soil moisture-precipitation and soil moisture-temperature feedback. Many regional processes that were modified due to increased SNAO could mask the 20 significant relationship among precipitation, temperature and tree growth. These processes can have a profound impact on tree growth, since incoming solar radiation and moderate heat flux are supporting factors for growth (Weitzenkamp et al., 2007). The weakening of the correlation between radial tree growth and climatic variables after 1970s (Fig. 5) may therefore be a consequence of changes in large-scale circulation 25 patterns, which affect local stability conditions and therefore many regional processes. Changes in incoming solar radiation due to increased cloud cover could weaken the temperature and precipitation growth relationships of P. nigra by having a direct effect on gross primary production. Identification of significant processes is beyond the scope of this article but should be addressed in future studies. Such variation or weakening of the summer temperature/precipitation signal in SUMMER group is not seen in the SPRING group. Analysis of cyclone activity over the Adriatic Sea has revealed that more pronounced changes in cyclone frequency occur during the summer than in spring (Bartholy et al., 2009), which is thus in accordance with the observed signal in the SPRING group. Evidence of change in the climate-growth relationships for P. nigra is reported for almost the same time period for Austria (Leal et al., 2008) and the Eastern and Northern Iberian Peninsula for P. nigra, Pinus sylvestris (L.) and Pinus uncinata (Ramond ex. DC) (Andreu et al., 2007). The decrease in sensitivity of P. nigra trees in Austria is ascribed 10 to an improvement in water-use efficiency arising from a stimulation of photosynthesis and declining stomatal conductance as a consequence of the increasing CO 2 concentration in the atmosphere (Leal et al., 2008). On the other hand, Andreu et al. (2007) report that, at the end of the first half of the 20th century, the relationship between tree growth and late summer/autumn temperatures of the year before growth changed and 15 became stronger. Higher correlations between larch tree growth and summer temperature after the 1970s are also reported from the Maritime French Alps (Büntgen et al., 2012). Sarris et al. (2007), meanwhile, present evidence for the Eastern Mediterranean area of rapid mortality of trees due to lack of precipitation after the late 1970s. 20 Based on two P. nigra chronologies (from two groups), we reconstructed the two most significant climate factors -MJJA temperature for the period 1701-1901 (200 yr) and JFM temperature for the period 1685-1901 (216 yr). The strength of the model for reconstructing extreme cool or hot summer temperatures depends on the percentage of explained variability and user-defined thresholds for percentiles. Again, thresholds are 25 specified by the final user of the reconstruction and the higher percentile values that are chosen, the fewer (but more extreme) events will be discovered. Touchan et al. (2008) used the 12th and 88th percentiles as a threshold for dry and wet extreme events. correlation between reconstructed and observed precipitation was r = 0.79, which is stronger than ours (r = 0.58). Using the published threshold, we identified only 4 predicted out of 17 observed hottest summers and 4 out of 17 observed coolest summers. We therefore adjusted the threshold to the 15th and 85th percentiles and re-applied the model to the reconstructed series and discovered additional extreme years. With 5 the new threshold, we could predict 6 out of 17 observed hottest summers and 6 out of 17 coolest summers (Fig. 7). When using the threshold of the 15th and 85th percentiles on reconstructed MJJA temperature in the period 1701-1901, we discovered 15 cool and 16 hot summers (Fig. 8). Reconstructed extreme hot and cool summers in BiH were compared to those published in the literature for other regions (Table 4).

Identification of extreme periods
We also tested the 12th and 88th percentiles (Touchan et al., 2008) for identification of extreme spring events. Using the 12th and 88th percentiles, we discovered 4 cold and 1 warm spring, while with our threshold, set at the 15th and 85th percentiles, we were able to reconstruct 2 out of 17 observed warm and 5 out of 16 observed cold springs in the chosen meteorological data set. Years 1917Years , 1940Years , 1963 were recognized as 15 years with cold JFM period in BiH; this years had also anomalously cold January-April seasons over much of Europe and the Mediterranean area (Luterbacher et al., 2010). Due to the poorer strength of the relationship between temperature and TRW indices in the SPRING group, the number of extreme JFM events is lower than with MJJA. Comparing the identified hot and cool summers to reported extreme events from the 20 surrounding regions, some common events between the Eastern Mediterranean and Continental Europe were observable. For example, the summer of 1725 was the hottest in our reconstruction and it was part of a two-yr major drought in Anatolia and Syria (Touchan et al., 2007). Similarly, there is a connection with Slovakia and the Czech Republic, since our predicted cool summer in 1871 coincides with a climatic event in 25 Slovakia (Büntgen et al., 2010b). In general, cool summers (1984,1896,1871,1831,1735) are related to oscillation patterns from Continental Europe and regions north of BiH, while warm events (1802,1782,1725,1715) are more related to oscillation from the South-Eastern/Eastern Mediterranean ( by the strong and consistent anti-phase relationship between a summer temperature reconstruction from Bulgaria and a reconstruction of the sNAO, which suggests that the sNAO pattern is the main driving force of the teleconnection between summer temperatures in South-Eastern versus North-Western Europe (Trouet et al., 2012). Comparing years with extreme hot summers to years with extreme cold springs 5 (since both hot summers and cold springs have a negative effect on radial growth), we found that 1725 and 1782 were extreme in both groups. Cool and wet conditions in the spring of 1725 prevailed over BiH and Slovakia (Brázdil et al., 2008). Later, during the summer of the same year, extensive anomalous low-pressure conditions, extending from Northern to Central Europe, may have been connected with the wet and cool 10 conditions over Slovakia (Brázdil et al., 2008), while BiH was influenced by the hot summer weather from the Eastern Mediterranean, similar to Anatolia and Syria (Touchan et al., 2007). The winter of 1782, also found in BiH, is reported for Greece as harsh and cold, with Lake Karla freezing and the destruction of olive and fruit trees, as well as the death of animals, while plague and the deaths of people are mentioned for BiH 15 (Xoplaki et al., 2001). Later in the year, drought events during the summer season extended north to Slovakia (Büntgen et al., 2010a), east to the Western Black Sea region of Turkey (Akkemik et al., 2005) and south-west to BiH. The winter of 1929 is also interesting, characterized as an extremely cold winter in Slovakia (Büntgen et al., 2011). In this year, the sea froze in the Venice lagoon (Camuffo 20 and Enzi, 1992). The average monthly January-March temperature in 1929 for BiH was −3.9 • C, while the long term average   closer examination of the discovered climatic events from BiH, further investigation of the documentary archives with an emphasis on the region of the Western Balkans is needed.

Conclusions
The mixed climate response divided 7 site chronologies into two groups. Tree-ring 5 widths of P. nigra from the sites Blace, Peručica, Šator and Konjuh (SPRING group) contain a valuable and significant January-March temperature signal, while tree-ring widths from the sites Krivaja, Prusac, Šipovo (SUMMER group) contain May-August temperature and precipitation signals. Both signals from the SUMMER group are stable from the beginning of the available CRU TS3.1 climate data until the late 1960s. In the 10 period 1970-1990, trees had a weaker response to summer climate. We associate this with circulation patterns in the Adriatic Sea, where, according to the literature, the number of summer cyclones in the central part of the Adriatic Sea increased significantly after 1970, affecting precipitation distribution and abundance in the Western Balkans and, consequently, also the P. nigra response to climate. We proved that mean May- 15 August temperature is the main factor affecting the growth of the trees of the SUMMER group and January-March temperature from the SPRING group. By means of spring and summer temperature reconstruction and the application of the percentile method for identifying extreme climatic events, we discovered 15 cool and 16 hot summers and 3 cold and 3 warm springs. The identified cool summer events from the BiH area 20 are related to oscillation patterns in Northern and Continental Europe, while hot summer events are more related to oscillation in the South-Eastern/Eastern Mediterranean. More thorough identification of these relationships should be a part of future studies. Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Institute of Bosnia and Herzegovina for supplying us with the meteorological data used in this study. We thank Maarten De Groot for advice on statistical analysis of the data. We also thank Špela Jagodic and Robert Krajnc for their valuable assistance in the field and laboratory. This work was supported by the Slovenian Research Agency through the program and research group "Forest, biology, ecology and technology" P4-0107 and a doctoral study grant for Simon Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Fig. 9. Comparison of predicted (black line) and observed January-March mean temperature (grey line) with black circles marking warm and cold springs as defined by 15th and 85th percentile thresholds of the measured data (dashed horizontal lines). EPS marks the end of usable length of reconstruction.