The impacts of climate change on society are most strongly felt through the increasing intensity and frequency of extreme events (e.g., floods, drought, heat waves, and/or cold spells)1,2,3,4,5,6,7,8,9,10,11,12. The economic losses caused by weather- and climate-related events in Europe increased from 7.4 billion Euro in the 1980s to 13 billion Euro over the period 2010–201713. During the 2003 European heatwave ~70,000 heat-related deaths were recorded14 and >17 billion Euro of direct economic losses were incurred13. Climate projections indicate that the frequency and magnitude of extreme events are expected to increase in a warming world15, but assessing projections against observational data is hampered by the relatively short instrumental records available. In particular, trends based on short records are very sensitive to the beginning and end dates and do not, in general, reflect long-term climate trend16. Natural paleoclimate archives (e.g., corals, ice cores, shells, sediment records) have become an important tool to supplement instrumental records and tree rings are one of the most valuable17, due to their annual resolution, precise dating, and widespread occurrence. Tree ring chronologies can be thousands of years long and allow past climate to be explored using a range of paleoclimate proxies, such as tree ring width, maximum density, and stable isotopic composition of the wood or its components18,19. Stable oxygen isotopes have the advantage that the processes of fractionation are well understood and they can record paleoclimate information in areas where other tree-rings proxies do not register the climate signal20,21,22,23. Many dendroclimatological studies focused on reconstructing pre-defined teleconnection indices like the North Atlantic Oscillation24, El Niño-Southern Oscillation25, Atlantic Multidecadal Oscillation26 or on the reconstruction of regional temperature and/or precipitation27,28. So far, relatively few studies have tried to link tree-ring-derived proxy signals to synoptic-scale pressure patterns21,23,29,30. However, it is the large-scale atmospheric circulation, through its prevailing high- and low-pressure systems, that strongly controls the frequency and magnitude of extreme events.

This study aims to develop the first annually-resolved tree ring δ18O time series for the Eastern Carpathians (Călimani Mountains), Romania (Fig. 1), in order to reconstruct the variability of the summer Standardized Precipitation-Evapotranspiration Index (SPEI) for the last ~700 years and to analyze the link between δ18O variability and large-scale atmospheric circulation. The palaeoclimate potential of stable isotopes in Pinus cembra L. (Swiss stone pine) tree-ring cellulose from the Călimani Mountains has been demonstrated recently30, showing that δ18O variability allows high-resolution paleoclimatic reconstructions over the eastern part of Europe, where few such reconstructions are available.

Fig. 1: Location of the investigation area.
figure 1

a Distribution of Pinus cembra in Europe99, and b topographic map of Romania showing the sampling site.

Results and discussion

Present-day climate

It has been long recognized that persistent high-pressure systems play an important role in the generation of heat waves, soil moisture deficit, and droughts both at global and regional scales10,31,32,33,34, and those long-lasting blocking situations associated with stationary Rossby waves are the leading drivers of droughts and heat waves1,35,36,37. To optimally analyze the link between summer large-scale atmospheric circulation and drought variability at the European level, over the observational record, we computed the coupled modes of variability between summer geopotential height at 500 mb (Z500)38 and August SPEI3 index, by employing a canonical correlation analysis (CCA). The first 10 empirical orthogonal functions (EOF) of Z500 summer (JJA) capture ~80% of the total variance, while the first 10 EOFs of August SPEI3 capture ~85% of the total variance. The optimal number of retained EOFs was chosen so that by adding or removing one EOF, the CCA would not change considerably39. The first coupled mode of variability between Z500 and SPEI3 reveals that a high-pressure system extending over south, central, and eastern Europe (Fig. 2a) corresponds with drought episodes over the same region (Fig. 2b), and a low-pressure system over the northern Atlantic and northern Europe is correlated with wet climatic conditions over northern Europe. The year-to-year variations of the normalized temporal components of the first CCA pairs are shown in Fig. 2c. The two time-series are significantly correlated (r = −0.91, p < 0.01). The negative correlations between standardized amplitudes corresponding to CCA1 Z500 and CCA1 August SPEI3 (Fig. 2c) confirm the significant relationship between drought variability at the European scale and large-scale atmospheric circulation. The two canonical time series present strong interannual variability, over the 1901–2015CE period, with the highest amplitudes (towards higher pressure and drought) over the 21st century. Overall, dry (wet) periods over the analyzed region are driven by the occurrence of a high (low) pressure system extending over most of the European continent and a low (high) pressure system over Iceland extending up to Fennoscandia.

Fig. 2: Coupled modes of variability.
figure 2

The first coupled mode of variability (CCA1) between a summer Z500 and b August SPEI3; c the standardized amplitudes corresponding to CCA1-Z500 (blue line) and CCA-SPEI3 (bars).

δ18O Chronology description

A stable oxygen isotope chronology, spanning 1331–2012CE, was constructed by measuring δ18O on the cellulose of individual tree rings using 16 trees (five living trees and eleven dead wood samples) (Fig. 3a and Supplementary Figure 1). The shortest and longest analyzed sections contained 90 and 370 tree rings, respectively (Table 1). Replication varies between two and five trees, with a two-replication record between 1331 and 1372CE, and the highest replication for the 20th century, while for most of the chronology the replication is four (Fig. 3). The mean of the δ18O raw values is 28.8‰ and the data ranges between 26.7‰ in 1837CE and 31.6‰ in 1946CE. The δ18O chronology presents periods with low δ18O values: 1428–1466, 1644–1703, 1800–1823, and 1967–1988, and periods with high δ18O values: 1470–1487, 1621–1636, 1704–1720, 1890–1962, and 1989–2012. Subsample Signal Strength (SSS) varies between 0.85 and 0.95 for the analyzed period, while the mean inter-series correlation (Rbar) and the Expressed Population Signal (EPS) values for the 1331–2012CE period range between 0.51 and 0.93, and between 0.80 and 0.98, respectively, for different time windows (Fig. 3), indicating that a strong common signal is preserved40,41. The linear reconstruction model (r = −0.71, p < 0.001) is thus strong and temporally stable, passing all of the conventional verification tests (Supplementary Figure 2).

Fig. 3: Raw δ18O values and the corresponding statistics.
figure 3

a The raw δ18O values measured in the Pinus cembra L. from Călimani Mountains, Romania (blue line), 31-year running mean (black line), and 95% calculated confidence level of the raw δ18O values (pink line). b The inter-series correlation (Rbar) values calculated for the raw δ18O series with different running window: 50 years with an overlap of 25 years (green line), 40 years with 20 years overlap (red line), and 30 years with 15 years overlap (blue line). c The Expressed Population Signal (EPS) calculated with different running window: 50 years with an overlap of 25 years (green line), 40 years with 20 years overlap (red line), and 30 years with 15 years overlap (blue line). d Subsample Signal Strength (SSS) for the raw δ18O series (green line). e Replication (dark blue line). The gray bars represent time periods for which the replication is ≤2.

Table 1 The analyzed trees and tree ring chronology length.

The δ18O values of tree-ring cellulose are mainly affected by the isotopic ratio of precipitation, which varies in response to the atmospheric pressure patterns during the summer42. Further fractionation occurs in the leaf, driven mainly by the difference between internal and external water vapor pressures, and thus by relative humidity19. Dry summer conditions thus lead to higher δ18O values in leaf sugars and therefore wood cellulose, so that, a negative correlation between tree ring δ18O and the SPEI index is expected.

Climate reconstruction

The reconstruction of August SPEI, for an accumulation period of 3 months (SPEI3) (Supplementary Figure 2), extends back to 1331CE (Fig. 4a). Variations of the δ18O values capture the August SPEI3 changes both at high and low frequencies (from interannual to multidecadal scales). For the reconstructed period, 44 years of extremely low and 45 years of extremely high SPEI3 values were found (Table 2, Supplementary Tables 1 and 2). However, their temporal distribution is not uniform over time, and several distinct periods can be highlighted. The period 1331–1500CE is characterized by nine low extreme years, and seven high extreme years. The period between 1500 and 1850CE is characterized by predominantly high extreme years (33 years from a total of 45) and only 18 low extreme years. After 1850CE low extreme years dominate (17 negative extreme years from 44, and only five high extreme years). The last high extreme year was recorded in 1955CE, while over the 1990–2012CE period were recorded 7 extreme low years. The temporal distribution of the extreme years overlaps very well with the frequency of extreme events from other regions of Europe. The high frequency of the positive extreme events during the 1500–1850CE period overlaps with the Little Ice Age, characterized by reoccurring cold and wet climatic conditions in this region43,44. In contrast, the only three positive extreme events and a higher frequency of negative extremes in the most recent part of the chronology could be related to the global climate change signal15,16. Our reconstruction indicates that the 20th and early 21st centuries are unprecedented, in terms of dryness, since 1331CE (Fig. 4b). The extreme events reported here correspond with historical reports of catastrophic events (Supplementary Tables 1 and 2). For example, the 1585CE drought is known as a very dry year, especially in spring, over the analyzed region. During the summer months, all the springs and pools dried and trees were killed by drought45,46. The year 1904, a negative extreme according to our reconstruction, was characterized by an extended drought in Moldova47. In Romania, it was so dry that the government had to take measures to avoid famine46. The 1904CE drought persisted for 18 months (from June 1903CE until the end of 1904CE). This year was reported as a dry year also in Switzerland and France45. The year 1946CE was one of the most severe droughts recorded in the 20th century and was characterized by very warm and dry conditions, with many absolute maximum temperature values greater than +40 °C in Moldova and Romania45,46,47. Historical sources have been used to analyze also positive extremes (wet years). For instance, 1593CE had a wet spring and summer, with a lot of rainfall and continuous floods46,47, not only in Romania but also in Ukraine45. In 1684CE catastrophic floods were reported over the analyzed region47, mainly due to four weeks of constant rain45. The year with the highest amplitude in our reconstruction is 1816CE, known as the “year without a summer”, with heavy rains, linked to the climatic aftermath of the eruption of Tambora45,48. Therefore, the robustness of our August SPEI3 reconstruction is validated by the good match of extreme years recorded by our reconstruction with historical sources for the last ~700 years (Supplementary Tables 1 and 2).

Fig. 4: Drought index reconstruction.
figure 4

a Reconstructed August SPEI3 (black line) for 1331–2012CE period, with the 31-year running mean (red line) and measured August SPEI3 for 1902–2012CE period (magenta line). Extreme dry/wet years are represented as lower-orange/upper-green triangles. Extreme years are defined when the SPEI3 index is lower than −1.5 or higher than +1.5 standard deviation. b Box plots of the reconstructed August SPEI3 for different centuries over the period 1331 (1301)–2012CE.

Table 2 The years corresponding to the low values (August SPEI3 < −1.5) and high values (August SPEI3 > 1.5) over the 1331–2012CE period.

The large-scale atmospheric circulation in a long-term context

Based on spatial correlation/stability maps, it appears that the spatial correlation pattern between δ18O values and SPEI3 over the last 110 years is stable in time and shows a clear dipole-like structure with negative and significant correlations over the central, southeast, and eastern parts of Europe, and positive and significant correlations over the northern and north-western parts of Europe30. Nagavciuc et al. (2020)30 have shown that during summer, high δ18O values in Swiss stone pine tree rings in the Călimani Mountains are associated with dry conditions over central, southeast, and eastern Europe, while low δ18O values are associated with wet summers over the eastern part of Europe. The dry and warm climatic conditions increase the contrast between ambient and intercellular vapor pressures, increasing the fractionation of oxygen isotopes in the foliage which transferred to the synthetized cellulose and archived in tree rings20,49, as well as changing the isotope ratios in precipitation. Overall, the strong correlation obtained for August SPEI3, explains 50.4% of local drought variability over the eastern part of Europe. Although the strong relationship between δ18O variability in tree rings from the eastern part of Europe, and large-scale atmospheric circulation has been reported21,30, this relationship was analyzed only over the observational record, covering the 20th century. In order to examine how large-scale atmospheric circulation can explain long-term drought variability, we computed composite maps using summer northern hemisphere geopotential height at 500 mb (Z500) from the 20th century reanalysis data over the period 1836–2012CE (Supplementary Figure 3) as well as by using paleoclimate reanalysis data50 for the period 1604–2003CE (Fig. 5). The composite map analyses based on both datasets show similar results. Wet years (SPEI3 values > +1 standard deviation) are associated with a low-pressure center over the central and eastern part of Europe and a high-pressure system over the northern Atlantic Ocean and northern Europe (Fig. 5a), while the dry years (SPEI3 values < −1 standard deviation) are associated with a high-pressure center over the central and eastern part of Europe and a low-pressure system over the northern Atlantic Ocean and northern Europe (Fig. 5b). These results are in agreement with the results reported by Nagavciuc et al. (2020)30. During summer, the low-pressure center over the central and eastern part of Europe is associated with a cyclonic circulation, which facilitates the advection of moist air from the Mediterranean region towards Romania, leading to above normal precipitation over this region (Fig. 5a)35,51. The high-pressure center over the central and eastern parts of Europe is associated with anticyclonic circulation, which favors the advection of dry and warm air from the northern part of Africa toward the south-eastern part of Europe thus generating very dry and hot summers over the study site (Fig. 5b).

Fig. 5: Large-scale atmospheric circulation.
figure 5

a The composite map between High SPEI3 (< −1 std. dev.) and summer (JJA) Geopotential Height at 500 mb (Z500)100 and b the composite map between Low SPEI3 (>1 std. dev.) and summer (JJA) Geopotential Height at 500 mb (Z500). The hatching highlights significant anomalies at a confidence level of 95%. Units: Z500 (m). Analyzed period: 1604–2003CE.

Stability of the drought signal in the reconstruction

The relationship between large-scale atmospheric circulation and European climate has been found to be non-stationary in time, especially when pre-defined teleconnection indices are used32,52,53. In order to overcome the problem of non-stationarity and to test if the identified relationship between the reconstructed SPEI3 index and the large-scale atmospheric circulation is stable over time, we employed a methodology used for the monthly to seasonal prediction of the mean runoff of the Elbe River51,54, but also applied in dendroclimatological studies30,55. Over the last 400 years the relationship between the reconstructed SPEI3 index and summer (JJA) Z500 is stable, showing a significant negative correlation over a large part of eastern Europe (Supplementary Figure 4). Based on the stability map (Supplementary Figure 4), we have defined a Z500 index averaged over the central and eastern part of Europe (black square in Supplementary Figure 4) to place the evolution of the increased high pressure over this region in a long-term context (Fig. 6). Over the last 400 years, prolonged dry (wet) years were associated with positive (negative) Z500 anomalies over the central and eastern parts of Europe. Although until 1850CE, both time series show no significant change in their temporal evolution, starting with ~1850 both time series present a significant (99% significance level) negative (SPEI3) and positive (Z500) trend. The 31-year running mean of the SPEI3 intersects with summer Z500 around 1860 CE, after which their opposing trends continue until the present (Fig. 6c). From a statistical perspective56, the year 1860 CE appears as a change point in both time series (Fig. 6c). The values of SPEI3 and summer Z500 index are both unprecedented over the 20th and 21st centuries. The box plot of August SPEI3 index and Z500 index over two distinct periods 1604–1850 and 1851–2002 CE (Supplementary Figure 5) indicates that there is a significant difference (p 0.001, based on a two-test sample, Kolmogorov and Smirnov) in the distribution of the two-time series over these two time periods. The 1604–1850 CE period is characterized by a higher frequency of wet years and lower Z500 values, while the period 1851–2003 CE is characterized by an increase in dry years and a shift towards higher values of the Z500 index. A similar abrupt shift in the values of δ18O, at ~1850 CE, has been captured in a coral record from the northern Red Sea57. The shift in the δ18O coral record was interpreted as being driven by a reorganization of the large-scale atmospheric circulation, which is in agreement with our findings. Moreover, at a decadal time scale (e.g., after applying a running mean of 31 years) the tree-ring δ18O record correlates with the coral δ18O time series from the northern Red Sea at the 99% significance level (r = −0.80).

Fig. 6: Drought reconstruction and large-scale drivers.
figure 6

a Reconstructed SPEI 3 (green line) and 31-year running mean (orange line); b summer (JJA) Z500 time series averaged over the stable region: 13.5°E–35.0°E, 30°N–52°N (blue line) and 31-year running mean (red line). c The direct comparison between the 31-year running mean for SPEI 3 reconstruction (orange line) and for summer Z500 time series (red line) and d Occurrence of positive Z500 anomalies and dry events (Z50075/SPEI325). Z50075/SPEI325 indicates that we took into account the common years when the Z500 anomalies were higher than the 75th percentile and SPEI3 was smaller that the 25th percentile.

To further emphasize the influence of Z500 anomalies on the variability of the reconstructed August SPEI3, we also computed the changes in the occurrence of concurrent extremes (e.g., positive Z500 anomalies/low SPEI3) over the last 400 years. Throughout the analyzed period, drought events were accompanied by positive Z500 anomalies, with an increased frequency after 1850 CE (Fig. 6d).

Drought at European scale

To have a long-term perspective of our reconstruction in comparison with other hydroclimatic reconstructions available at European scale and to assess their spatial synchronicity, our reconstructed August SPEI3 time series was compared with other reconstructions available at European scale. In a first step we compared our August SPEI3 time-series with the Old World Drought Atlas (OWDA)58 averaged over the region 25°E–28°E, 46°N–48°N (Supplementary Figure 6). As it can be inferred from Supplementary Figure 6 there are important differences of variability and magnitude between our reconstructions and the OWDA-based time-series. Overall, the OWDA does not capture low-frequency drought variability and does not show any trend over the 20th and 21st centuries (Supplementary Figure 6). The mean values of the OWDA reconstruction are below 0 for the whole analyzed period, with the minimum (driest) during the 17th century and the maximum during the 20th century. In contrast, our August SPEI3 reconstruction, based on stable isotope data that has not been statistically de-trended, retains low-frequency variability, with the maximum August SPEI3 values (wet) during the 17th century and with the minimum in the 20th century, results that are in a good agreement with the historical evidence and more recent observations (Supplementary Tables 1 and 2). Our results emphasize that paleoclimate reconstructions focused on the western and northern parts of Europe cannot explain the complex spatial variability of climate over the entire continent, especially over the eastern part of Europe. For example, the recent reconstruction based on tree-rings stable isotopes from the Czech Republic and Southern Germany59, only shows strong correlations over a small region of central Europe (Fig. 3b in their paper). The strong offset between our reconstruction and the Büntgen et al. (2021)59 drought reconstruction (Supplementary Figure 7) emphasizes the need for an extended network of reconstructions covering the entire Europe. Overall, it is clear from Supplementary Figure 7 that the drought reconstruction from a small region like the Czech Republic and southern Germany are not able to capture fully the drought variability at European scale. This is further emphasized by comparing our reconstruction to drought history from the Czech Republic and southern Germany59, and a drought reconstruction covering the entire Germany60. Although the two reconstructions59,60 are coming from closely related regions, even over some overlapping ones (such as southern Germany), wet (dry) periods over Germany as recorded by August SPI360 are not matching with wet (dry) periods as recorded by the drought reconstruction from the Czech Republic and southern Germany (Supplementary Figure 7b and 7c). Similar results have been recently shown by Ionita et al. (2021)36, which have shown that the recent droughts of 2015 and 2018 are not unprecedented over the central part of Europe (with a focus on Germany), which is in disagreement with the conclusions of Büntgen et al. (2021)59 study and in agreement with the findings of Glaser and Kahle (2020)60. Therefore, a priority for a better understanding of climatic variability over Europe is high-resolution paleoclimate reconstructions in the Eastern part of Europe and not only.

Comparisons of our summer drought reconstruction and other reconstructions across Europe (Fig. 7 and Supplementary Figure 8) indicates the August SPEI3 is in agreement with reconstructions preserving similar characteristic of the spatial variability. The Călimani August SPEI3 record accords with the summer temperature reconstruction for the same site61 and with summer (JJA) SPEI reconstruction over France62 (Fig. 7). The lack of low-frequency variability in the other precipitation and drought reconstructions based on tree-ring width from the vicinity of the study area precludes a direct comparison with our August SPEI3 reconstruction, therefore we also computed the 31 years running mean for each reconstruction (Supplementary Figure 8i, j, k, l, m, n, o, p). Overall, the results showed that our reconstruction retains both high and low-frequency variability similar to other reconstructions; however, their synchronicity is not stable in time. The high-frequency similarities are expressed through the extreme years registered by different reconstructions. On a decadal time scale, our August SPEI3 record accords with reconstructions from the North Aegean63, Czech Republic64,65,66, southwestern Turkey67, Eastern Mediterranean68, and from the south-eastern part of Romania69 (Supplementary Figure 8). Decadal-scale variability of extreme climatic conditions also appears robust. For example, the 1470–1490CE period represents one of the most severe drought periods, recognized in numerous studies over Europe70,71. The drought conditions in this period were severe across Europe and can also be observed in the OWDA drought reconstruction (Fig. 7). This period is also characterized by high-temperature anomalies in the study area, as well as over the central part of Europe (Fig. 7). Another example of the severe long-term drought condition prevailed from 1700 to 1730CE is in the opposite phase with the drought conditions over Scandinavia72, emphasizing the dipole-like structure in the drought conditions over Europe73. This dipole structure is even more obvious in the last decades when southern and central-eastern Europe show an increase in the drought severity and frequency, while the SPEI reconstructions for northern and western Europe show a tendency towards wetter conditions. This dipole structure can be also seen when looking at wet periods in the reconstructed SPEI3 indices (Fig. 7). For example, the 1585–1615CE and 1800–1830CE periods are characterized by wet climatic conditions and high-temperature anomalies over southern central, and eastern Europe, while over the northern part of Europe dry conditions prevailed.

Fig. 7: Comparison among different reconstructions at European level.
figure 7

a Reconstructed August SPEI3 based on the δ18O in tree ring cellulose from Călimani Mts. from 1331 to 2012 period (gray line). b Summer (JJA) SPEI reconstruction over France (magenta line)62 over the period 1326–2004CE. c Summer (JJA) SPEI2 reconstruction over Scandinavia (green line)72 for the period 1272–1995CE. d Regional mean Old World Drought Atlas (OWDA) drought index averaged over the eastern part of Europe (25°E−28°E, 46°N−48°N, green line) for the period 1252–2012CE58. e Summer temperature reconstruction over Romania (black line)61. f Time series of annual total solar irradiance (light gray)101 and g Aerosol optical depth (AOD, orange bars)102. The bold red lines in (af) represent the 31 years running mean for each time series.


In this study, we report the first reconstruction of the summer (June to August) Standardized Precipitation-Evapotranspiration Index (SPEI3), 1331–2012CE, for eastern Europe, based on annually-resolved stable oxygen isotope ratios (δ18O) from Pinus cembra L. tree-ring cellulose from the Călimani Mountains, Romania. Variation of δ18O captures the SPEI3 changes at both high and low frequencies, revealing changes in the prevailing large-scale atmospheric circulation over central Europe. At the end of the Little Ice Age (~1850CE), there was a shift in the large-scale atmospheric circulation towards higher pressure systems over the central and eastern parts of Europe, which led to a drying tendency over the eastern part of Europe over the 20th and 21st centuries, which to the best of our knowledge is unprecedented over the last ~700 years (Fig. 2b). This is the first and longest reconstruction of drought variability for eastern Europe based on stable oxygen isotopes in tree-ring cellulose, helping to fill a missing gap in the spatial coverage of paleoclimate reconstructions.


Study site and local climate

The Călimani Mountains belong to a volcanic chain, situated in the north-western part of the central Eastern Carpathians (Fig. 1). With the highest peaks exceeding 2000 m, the region experiences a mountain temperate-continental climate, with severely cold winters and cool summers. The mean maximum annual temperature over the analyzed region is 8.5 °C and ranges between −2.7 °C in January and +18.8 °C in July–August74. Mean annual precipitation is 889 mm, with a summer peak in June–July when amount exceeds 100 mm/month74. During winter, thick snow cover can last for up to 200 days. Forests are dominated by Swiss stone pine (Pinus cembra L.) mixed with Norway spruce (Picea abies Karst., L.) which are replaced by mountain pine (Pinus mugo) towards higher elevations. Since 1975 the area has been protected due to natural forests with old trees and abundant subfossil wood. Open-pit sulfur mining was operated nearby between 1965 and 1992CE, with the most intense activity during 1974–1986CE75 disturbing the climate-growth relationship and leaving clear signals also in the chemical composition of the wood76.

Sample collection, preparation, and stable isotope measurements

The Swiss stone pine samples were collected in autumn of 2012, from Călimani Mountains (Călimani National Park) (Fig. 1), from elevations ranging from 1450 m a.s.l. to 1850 m a.s.l. (treeline) while the current timberline is situated at 1700 m a.s.l.77. In total, 16 cores (one core per tree) were used for stable isotopic measurements, out of which five were living trees (labeled Tree 1 to 5) and eleven were subfossil trees (labeled Tree 6 to 16). From the living trees, samples were collected using an 11 mm diameter increment borer, while for the subfossil trunks found on the forest floor, a sample disk was sawn, and the trunks were left in the original place. Tree-ring widths (TRW) of samples were measured using LINTAB equipment and TSAP 0.53 software, with 0.001 mm accuracy for accurate cross-dating against the local master chronology61 and checked for missing rings or dating accuracy with COFECHA software78. The analyses of stable oxygen isotopes were performed for the 1331–2012CE period (Fig. 3). The tree ring samples were split with a scalpel to obtain annual resolution before measurement of each ring individually (no pooling). Alpha-cellulose, extracted using the modified Jayme-Wise method79,80, was homogenized by a standard ultrasonic protocol81 using VCX130 (Sonics & Materials Inc/USA) device and dried at 70 °C for 24 h. For the measurements of oxygen isotope ratios (δ18O), 0.2 mg (±10%) of α-cellulose was packed in silver foil and pyrolyzed over glassy carbon at 1450 °C using a ThermoQuest TCEA interfaced with a Thermo Delta V Advantage IRMS82,83. All isotope values are reported in per mil (‰) relative to the Vienna Standard Mean Ocean Water (VSMOW)84, using the traditional δ (delta) notation. The analytical precision of the measurements was better than 0.1‰. All samples were measured in triplicates, but if their standard deviation exceeded 0.2‰ two additional measurements were performed to check potential outliers and reduce uncertainity. The individual δ18O series were tested systematically for trend effect during the juvenile years in a separate study30,85. Since they do not show any age trend, the raw δ18O can be used for further paleoclimatic reconstruction. In order to minimize any effect of offset between individual time series, the method described by the Nakatsuka et al. (2020)86, was applied to combine the individual time series into a composite δ18O chronology. The Expressed Population Signal (EPS) and the inter-series correlation (Rbar) values were calculated for the δ18O series for the 1331–2012CE period, with different running windows: 50 years with an overlap of 25 years, 40 years with 20 years overlap, and 30 years with 15 years overlap. Also, Subsample Signal Strength (SSS) was calculated for the same period. We do not combine oxygen and carbon isotope results because the mechanistic controls on fractionation are different19 and they do not carry the same climate signal29.

Climate data and statistical methods

The potential of climate signals in oxygen isotope ratios of Swiss stone pine tree-ring cellulose from the Călimani Mountains was tested in a separate paper30. According to the results reported30, the highest correlation coefficients and best statistical skill were obtained for δ18O values and relative humidity at the local scale, and δ18O and the summer SPEI drought index at the European scale. The correlation between δ18O chronology and the SPEI index was tested for different months and different periods of accumulation (from one month to 12 months), and the highest coefficient of correlation was obtained for SPEI3 August (Supplementary Figure 9). As such, the SPEI3 August drought index was used to provide a long record of summer drought conditions over the eastern part of Europe. For the reconstruction model we used the Standardized Precipitation Evapotranspiration Index (SPEI) covering 1902–2012CE. SPEI was computed based on the monthly precipitation (PP), monthly mean air temperature (TT), and the potential evapotranspiration (PET) data from the CRU T.S. 4.04 dataset74 dataset using the R-package SPEI ( and has a spatial resolution of 0.5° × 0.5°.

Calibration and verification procedure

The reconstruction model of the drought index was developed using the R packages dplR87 and treeclim88, using the linear regression method. To test the predictive skills of the model for drought index reconstruction, the data were split into two equal periods: calibration and verification (Supplementary Figure 2). The reduction of error (RE), coefficient of efficiency (CE), and R2 statistics were used to assess the skill of the regression model89. The Durbin–Watson statistic (DW) was used to test the trend in the residuals90. The calibration and verification models passed all of the conventional verification tests, both in the forward and the reverse modes. For the August SPEI3 model, the RE = 0.63, CE = 0.63 and DW = 1.82 for the forward model, and RE = 0.43, CE = 0.42 and DW = 1.83 for the reverse model. The positive values of the RE and CE indicate good predictive skills of the model, which suggest that the reconstruction model is robust and stable over time. The DW test values indicate that the residuals of the SPEI3 model have low autocorrelation. The results of the statistical tests indicate that the reconstruction model was reliable and can be considered for the reconstruction of August SPEI3 at local and at the European scale. When a linear regression model is used for climate reconstruction, the variance underestimates that of the target data91,92, so the reconstructed August SPEI3 values were rescaled with mean zero and one standard deviation.

Canonical correlation analysis (CCA)

To identify the coupled patterns of variability between the summer large-scale atmospheric circulation (Z500) and August SPEI3, we used a Canonical Correlation Analysis (CCA). CCA is a powerful multivariate technique used to identify pairs of patterns with the maximum correlation between their associated time series39. Before applying a CCA, the dimensionality of August SPEI3 and geopotential height at 500 mb (Z500) datasets were reduced by an Empirical Orthogonal Function (EOF) analysis. The first 10 EOFs of JJA Z500 and August SPEI3 were retained as input in the CCA. The first10 EOFs of Z500 JJA capture ~80% of the total variance, while the first 10 EOFs of August SPEI3 capture ~85% of the total variance. The optimal number of retained EOFs was chosen in such a way that by adding or removing one EOF, the CCA would not change considerably39. Among other statistical methods, CCA has the advantage of selecting pairs of optimally correlated spatial patterns, which may lead to a physical interpretation of the mechanism controlling the climate variability.

Stability maps

To test the stability of the relationship between the reconstructed drought index and the summer geopotential height, we make use of stability maps, a methodology successfully used in the seasonal forecast of the European rivers and Antarctic sea ice to examine the stationarity of the long-term relationship between our proxies and the gridded climate data54,93,94. In order to detect stable predictors, the variability of the correlation between the tree-ring parameters and the gridded data is investigated within a 51-year moving window over the 1604–2003CE period. The correlation is considered stable for those regions where the tree-ring parameter and the gridded data are significantly correlated at the 90 or 80% level for more than 80% of the moving window. The basic idea of this methodology is to identify regions with stable correlations (meaning the correlation does not change over time) between August SPEI3 and summer Z500.

Compound events

To test the influence of Z500 on the probability of occurrence of dry events, we employ a joint frequency analysis of compound events95 (e.g., the co-occurrence of positive Z500 anomalies and dry events)96. For each time series (e.g., Z500 and SPEI3 August) we computed a binary variable (Y = 1 for co-occurrence and Y = 0 for non-occurrence) which indicates the occurrence based on Z500 and SPEI3. For the specific threshold of the variables, the occurrence of a compound event can be expressed as:

$$Y=\left\{\begin{array}{c}1,\,P\le {px},\,T \, > \, {tx}\\ 0,\hfill\,{{{{{{\rm{otherwise}}}}}}}\end{array}\right.$$

where px indicates the precipitation threshold and tx indicates the temperature threshold, for example. For the current analysis, we have chosen as threshold the 75th percentile for Z500 and the 25th percentile for SPEI3 August.

Sources of uncertainty

Paleoclimatic reconstructions give us the opportunity to explore climatic variability from times for which we do not have measured data, but every reconstruction contains a range of uncertainties. The sources of uncertainty are numerous and can occur at every step of the process97. The use of wood samples from the same site removed uncertainties arising from different wood sources or possible differences in the registered signal. Measuring the isotope ratio on individual rings, rather than pooling98, and measuring at least three times, although expensive, ensures high precision, necessarily balanced by low replication. The reconstruction is based on three to five tree replications, and only for the period 1331–1373CE the replication is only two (Fig. 3). The potential offset between individual time series was removed by using the method described by Nakatsuka et al. (2020)86, to combine the individual time series into a composite chronology. The statistical properties of the composite chronology are good and the absence of any statistical de-trending ensures that it is possible to retain climatic information at all frequencies. In summary, our approach favors precision over replication, so the main source of uncertainty lies in the limited replication. The chronology provides, however, a firm foundation to which more data can be added in the future, allowing the magnitude of uncertainty to be both reduced and quantified more precisely.