Introduction

Europe is projected to experience direct and indirect impacts of continuing anthropogenic warming, particularly related to the hydrological cycle, such as droughts and floods1,2. Severe drought conditions in the summers of 20033, 2010, 20154 and 2018 demonstrated the damaging and expensive impacts of widespread European droughts5. Flooding caused by extreme precipitation in 2021 showed the harmful nature and devastating impacts of hydroclimatic extremes on society. With anthropogenic warming projected to exacerbate the occurrence of hydrological extremes around the world6,7, Europe faces large uncertainties in risks to its population, ecosystems and economies8,9. Europe’s vulnerability to hydroclimate extremes is further confounded by the complex interplay of spatially opposing trends10, a large degree of internal variability11 and brevity of long-term spatiotemporal records.

The most recent European summer drought (2015–2019) has sparked a debate as to whether it is within the range of natural variability12 or related to anthropogenic warming13. Long-term observations suggest that there had not been two consecutive summer droughts in central Europe in 250 years until 2018–201914. Based on traditional tree-ring records and reanalysis products, the recent drought is well within the range of natural variability thus not unprecedented12. By contrast, a long tree-ring isotope record from the Czech Republic indicates that the recent consecutive drought was unprecedented in the last two millennia13. It remains unclear whether this recent drought is part of natural climate variability or potentially a consequence of anthropogenic climate change. Obtaining a precise picture of the natural variability of the European hydroclimate is a challenge due to the sparse spatial coverage of moisture-sensitive proxy data15.

To understand the spatial complexity of the large-scale European hydroclimate, spatially explicit, long-term data sets are needed. Multi-century climate data are essential for the validation of climate models including comparisons to other paleoclimate, historical, and archaeological data sets. Tree-ring widths and maximum latewood density provide the longest and most replicated high-resolution records used in previous reconstructions of Holocene past climate15. However, tree-ring widths require statistical detrending, making it difficult to use them to reconstruct some aspects of low-frequency climate variability16. Furthermore, tree-ring width chronologies from European lowlands can display weak and ambiguous climate signals. In contrast, tree-ring stable isotopes are considered to be a more powerful proxy as they potentially require less statistical data treatment, and often exhibit clearer climate signals particularly in the temperate European lowlands13,17,18.

In this study, we reconstruct European hydroclimate based on a network of tree-ring stable isotopes of oxygen and carbon ratios. The network combines up to 400-year long, annually resolved records of deciduous oaks from European lowlands and conifers from boreal and mountainous sites. This is the first time that stable isotope records from tree rings, with their so far untapped high climate reconstruction potential, are used to obtain a gridded spatial reconstruction of the European summer hydroclimate.

Results

European tree-ring isotope network

The tree-ring isotope network is composed of 26 individual, well-distributed sites from across Europe (Fig. 1a). Each location has a stable isotope record of carbon (δ13C) and oxygen (δ18O) derived from tree-ring cellulose obtained from old living trees using standard procedures of dendrochronological dating and cross-dating16 and stable isotope analyses19 (Methods). The network is composed of eight deciduous broadleaf oak stands (Quercus spp.) and 18 coniferous stands (Pinus, Juniperus, Larix, Cedrus) ranging from elevations of 10–2200 m (Supplementary Table 1). Most of the broadleaf stands are concentrated in central-western Europe while the coniferous stands are mainly located at high latitudes and high elevations. The common period of all carbon and oxygen isotope records is 1850–1998 (Fig. 1b). Mass spectrometry of stable isotope ratios was conducted using pooled alpha-cellulose samples from at least four trees per site19. Coniferous species were measured using whole tree-ring wood material, whereas cellulose extraction for most oak sites was achieved on latewood only20. Previous studies based on the isotope network20,21 and on single sites have revealed common climate signals across Europe22,23,24,25,26,27,28,29, enabling a pooling of isotopic records derived from broadleaf trees and conifers into a single dataset.

Fig. 1: Overview of the European tree-ring isotope network.
figure 1

a Spatial distribution of the tree sites across Europe with annually resolved isotope records of δ18O and δ13C. b Temporal distribution of δ18O and δ13C records as bars. c–e Seasonal response of carbon, oxygen and tree-ring records62, respectively, with SPEI-2. Shown is the highest leading seasonal correlation with SPEI. Over the 1901–1998 period the highest correlation for each site is assigned to the closest gridboxes within 60 km.

Full size image

Seasonal climate signal of the tree-ring isotope network

Monthly correlation analysis with climate variables reveals a highly seasonal dependency of isotope records with summer conditions (JJA). While most δ18O records are positively correlated with summer temperature and anti-correlated with precipitation, δ13C records show the reverse climate response (Supplementary Figs. 1 and 2). Since both climate parameters are physically related and lower precipitation is mostly accompanied by high temperatures, it is difficult to separate out a specific forcing signal30. This strong covariability of temperature and precipitation suggests that drought variability is a more appropriate climate parameter to exploit the combined δ18O and δ13C climate information.

The majority of isotope sites across Europe display the highest seasonal correlations with the Standardised Precipitation-Evapotranspiration Index31 (SPEI) in summer (Fig. 1c, d). Independent of latitude, elevation or species, all δ18O records are most correlated with SPEI in summer (Fig. 1c). The only exception is the Turkish site (Isibeli), which shows the highest correlation with spring conditions23. Carbon records (Fig. 1d) exhibit similar seasonal correlations with predominately summer SPEI except for three sites (Vinuesa, Niepolomice and Col Du Zad). Across all sites, one or both isotopic records represent summer conditions during the current year of ring formation as a key influence on tree-ring isotope variability. Comparing the climate signals from isotopes with tree-ring widths shows substantial differences (Fig. 1e). Classical European tree-ring records32 mirror temperature and precipitation extremes but show diverse seasonal responses, with a greater influence of the cold season corresponding to greater continentality33. Even though a large proportion of European tree-ring width chronologies include a summer SPEI signal, there are spatial biases towards high-elevation sites such as the Alps or Pyrenees or climatically extreme sites in the far north or south. Lagged and carry-over effects that are present in many traditional tree-ring width chronologies limit the ability to retrieve a coherent continent-wide seasonal drought signal for Europe. The tree-ring isotope records promise a more homogeneous seasonal drought signal across the wide ranges of altitudes and latitudes and a better representation of low- to high-frequency variability (Supplementary Fig. 7), making them superior proxies for spatial reconstructions of hydroclimate variability compared to traditional tree-ring parameters15.

European hydroclimate reconstruction

We employ the well-established and tested point-by-point regression approach34 to spatially reconstruct summer hydroclimate variability (July–August SPEI) for the past 400 years. Similar to the climate field reconstruction efforts using tree-ring widths and maximum latewood density in Europe15 and other regions34,35, a nested and correlation-weighted regression approach is applied to split calibration (1914–1963) and verification (1964–1998) periods (Methods). The reconstruction explains up to 64% of summer SPEI variability in western-central Europe and approximately 30% averaged across the continent (Fig. 2a). Validation statistics demonstrate excellent agreement between instrumental and reconstructed SPEI at interannual and decadal timescales, with the strongest agreement in western, northern and parts of southern Europe (Fig. 2b). Most of the calibration and verification statistics indicate a significant and temporally stable climate-isotope relationship. The positive values of the reduction of error (RE) and the coefficient of efficiency (CE) at over 85% and 75% of all gridpoints (Fig. 2c, d), respectively, further highlight the robustness of the reconstructions. The skill of the reconstruction is strikingly good in parts of western and northern Europe, the skill is less in high-elevation regions such as the Carpathian Mountains, the Scandinavian Mountains and the Scottish Highlands, and in low-elevation regions such as north-east Germany and parts of Poland, potentially due to the lack of local tree-ring isotope sites, the regions being continental fringe locations, inconsistent climate-isotope relationships, or heterogeneous climate data (Supplementary Fig. 4). Long historical records and paleoclimate reconstructions13,15,36,37 further corroborate the SPEI reconstruction for the pre-instrumental period. There is high covariance between historical temperature records and the SPEI reconstruction on continental and regional scales, and high covariance between historical precipitation records and the SPEI reconstruction on local scales (Supplementary Table 2). Spatially resolved reconstructions of precipitation, PDSI and temperature are also highly correlated with our SPEI reconstruction (Supplementary Fig. 4). This provides additional assurance that our European summer SPEI reconstruction captures a large proportion of temperature and precipitation variability, is well verified during instrumental and pre-instrumental periods and is in parts, superior to existing reconstruction efforts.

Fig. 2: Spatial reconstruction statistics.
figure 2

Calibration and verification statistics for the SPEI reconstruction. Explained variance (R2) of a the calibration period (median CRSQ = 0.31) and b the verification period (median VRSQ = 0.22) on interannual and decadal timescales (insets) using a 5-year moving average filter. c Reduction of error of the calibration period (median RE = 0.20). d Coefficient of efficiency of the verification period (median CE = 0.15).

Full size image

European droughts and change points

The continental SPEI reconstruction shows variability on different timescales ranging from interannual to multi-decadal (Fig. 3a). Mostly wet conditions prevailed in the early seventeenth century exemplified by the first year of the reconstruction (1600), followed by generally drier conditions for almost two centuries. These generally drier conditions prevailed until the mid-nineteenth century, alternating with extreme wet conditions on an interdecadal scale. During this period, wet anomalies lasting several years, such as around 1840 or 1870, were rare. This is in contrast to the twentieth century, during which wet and dry phases commonly spanned 5 or more years, such as in the 1950s and 1960s, and there appeared to be greater decadal variability. The reconstruction reveals the spatial signature of well-known extreme years. For example, the significantly cold, wet summer of 1816, often referred to as ‘the year without summer’38, which corresponded with the Tambora eruption in Indonesia thought to have cooled the world by 2–3 °C39. The SPEI reconstruction (Fig. 3a, inset) reveals very wet conditions in western Europe during this particular year, yet those parts of western Europe (Spain) and northern Europe were not affected. Other large volcanic eruptions such as the Laki eruption (1773–1784) and the Krakatoa eruption (1883) show no coherent impact on the summer SPEI (Supplementary Fig. 5).

Fig. 3: Reconstructed European summer variability.
figure 3

a European field average of reconstructed SPEI (JA) from 1600 to 2018. A smoothed 13-year Chebyshev low-pass filter is shown as a black line. Insets show the spatial reconstructed SPEI for 1600, 1784, 1816 and 1883. b Change point analysis of European field average using the Wilcoxon rank-sum tests for sliding time intervals between 5- and 200-year time windows. The colour of a point corresponds to the tested time interval, and the year corresponds to the centre of the tested data. Significant change points are below the red horizontal line (p < 0.01). The Little Ice Age (≈ 1300–1860 CE) and the Maunder Minimum (1645–1715) periods are highlighted. Vertical lines indicate significant long-term (dark, >30 years) and short-term (light) change points. c Average conditions before and after significant long-term change points in 1651 (left) and 1874 (right).

Full size image

Change point analysis of the continental SPEI reconstruction reveals three distinct phases (Fig. 3b). The first phase, from 1600 to 1652, is characterised by generally a wetter climate, with anomalously wet conditions spanning multiple years. These pluvial years are most apparent in western-central and northwest Europe (Fig. 3c). Shortly after the start of the Maunder Minimum (1652), summer SPEI conditions shift to drier conditions, lasting over two centuries. This second phase is characterised by intense negative SPEI anomalies and a number of multi-year droughts. The Mediterranean, eastern and northern Europe are most affected by this change to drier summer conditions (Fig. 3c). By the end of the Little Ice Age, the third long-term change occurs and the third phase starts. Around 1875, after decades of generally drier conditions, the reconstructed SPEI changes significantly. Large parts of central and western Europe and the Mediterranean show milder summer conditions (Fig. 3c). These milder conditions prevail during the first half of the twentieth century until approximately the 1950s. In the second half of the twentieth century, we see a declining trend towards drier conditions, possibly linked to increasing temperatures, exemplified by the driest years on record (2003 and 2006) at the continental scale. These three distinct phases can, in part, also be found in other European reconstructions (Supplementary Fig. 6), but appear to be drought-specific features.

Recent trends and droughts

We use our SPEI reconstruction to place recently observed trends and droughts in the context of the past four centuries. Histograms summarise all 30- and 100-year linear trends in the European reconstruction (averaged across all gridpoints) from 1600 to 2018 (Fig. 4a, b). In the distributions of trends, we distinguish between pre-1970/1900 variability (grey), trends since 1970/1900 (light coloured) and the trend of the most recent 30 or 100 years (dark coloured). At a 30-year timescale (Fig. 4a), recent trends vary in sign, highlighting a large degree of variability. The most recent trend is a slight tendency towards drier conditions; however, it is indistinguishable from the distribution of historical trends derived from the entire reconstruction. At the continental scale, there is no significant 30-year drying trend. In contrast, considering a 100-year window, there is a clear tendency towards drier conditions (Fig. 4b). Trends starting after 1900, including the most recent trend ending in 2018, are in the lower quartile range. These overall trends towards drier conditions appear to be unusual, yet not unprecedented, considering the long-term context of the reconstruction. The spatial distributions of the recent trends (Fig. 4c, d) reveal that significant declines in the summer SPEI are mainly observed in southern Europe. In the latest 30-year period, parts of Spain and Italy experienced a significant decline in summer SPEI (Fig. 4c). When considering the trend over 100 years, this tendency towards drier conditions intensifies and expands to include large parts of southern, western and central Europe (Fig. 4d), including Spain, France, parts of Germany, Poland and Sweden. Fennoscandia is the only region that experienced a significant wetting trend in the past 100 years. This significant centennial wetting trend within the Arctic Circle contrasts sharply with the drying trends over the southern continental mainland. The spatial European drought reconstruction provides not only evidence of regional long-term trends but also helps to contextualise multi-year drought events such as the recent dry spells since 2015. Most of the European mainland experienced unprecedented drought conditions between 2015 and 2018 with moisture conditions very much below average and in some places the lowest on record (Fig. 4e). Large parts of central and western Europe, southern Spain and parts of Italy encountered this recent 4-year drought episode, unprecedented relative to both the instrumental record and the past four centuries. In contrast, during this period, parts of northern Europe including Scotland and parts of Fennoscandia, as well as southeastern regions, including Greece and Turkey experienced unusually wet conditions. These opposing conditions highlight the importance of spatial climate reconstructions.

Fig. 4: Recent trends and drought.
figure 4

a, b Contextualising recent 30- and 100-year trends of average European SPEI (JA). Histograms show the full range of the trends (grey shaded bars) prior to 1970 for 30-year periods (a) and 1900 for 100-year periods (b). Recent periods are shown in light red, except the most recent period, which is shown in dark red. Bar heights are normalised by the maximum occurrence. c, d Recent trends of gridded SPEI of the most recent 30-year period (c) and 100-year period (d) ending in 2018. Stippling indicates statistically significant trends (p < 0.05). e The recent drought period (2015–2018) is shown as deciles considering the full period (1600–2018).

Full size image

Our spatial reconstruction of summer drought conditions over Europe based on tree-ring isotope records provides evidence of historical changes and provides a unique long-term context for recent trends and drought events. We find that the most recent sequence of summer drought (2015–2018) is unprecedented in the past four centuries for large parts of Europe and is superimposed by long-term trends. This result corroborated and spatially extends the finding of the recently published Czech Republic summer reconstruction13. Our reconstruction also provides additional evidence of longer and possibly more severe droughts12. We find that during the Maunder Minimum and at the end of the Little Ice Age40, summer conditions across Europe were drier and multi-year drought episodes occurred frequently. This may have been due to the low solar activity at the time41, possibly similar to severe droughts during the Little Ice Age such as observed in the United States42. The observed tendency towards greater decadal SPEI variability and frequent multi-year pluvials following the end of the Little Ice Age43 could be related to the sudden reversal in Holocene temperature records44,45. To further confirm links between European drought conditions, solar forcing and anthropogenic warming, climate model simulations are currently being carried out within the CMIP6 framework, but the evaluation has not yet been finalised. At present, shortcomings of climate models to adequately represent hydroclimatic conditions hinder a direct comparison7,12,13,46,47,48,49,50. Our reconstruction is therefore providing a unique baseline for constraining past and recent changes and improving our understanding of droughts in a warming world.

Methods

Instrumental climate data

We use the Standardised Precipitation-Evapotranspiration Index (SPEI) as the instrumental target for our reconstruction. The SPEI is a multiscalar and multitemporal drought index which can quantify drought events on different timescales31. The global SPEI database51 ‘SPEIbase v2.6’ is calculated over the 1901–2018 period using precipitation and potential evapotranspiration data from the CRU TS4.03 dataset52. The potential evapotranspiration data is calculated based on CRU TS4.03 temperature, vapour pressure, cloud cover and wind field data, via the Penman–Monteith equation. On a spatial resolution of 1° × 1°, the SPEI combines the influence of precipitation and potential evapotranspiration to determine the accumulation of water deficit/surplus at different timescales ranging from short- (monthly) to medium-term moisture conditions and up to long-term (multi-annual) precipitation variability31.

Isotopic tree-ring records

We compiled a network of 26 individual tree sites of stable oxygen isotope (δ18O) ratios and carbon isotope (δ13C) ratios extracted from the corresponding tree-ring cellulose material. While the oxygen isotope records do not require any pre-processing, the carbon isotope records need to be adjusted to account for anthropogenic CO2 emissions of strongly depleted δ13C values. The use of the photosynthetic discrimination Δ13C takes into account the decreasing δ13C trend due to anthropogenic CO2 emissions. It removes this offset and eliminates this imposed anthropogenic trend. In contrast to the physiological age-related trend, which influences tree-ring widths, the δ13C records are influenced by the increasing anthropogenic trend of CO2 in the atmosphere53. Since industrialisation, the atmospheric CO2 concentration has increased from approximately 280 ppm to almost 415 ppm, a rise of almost 50%. The burning of fossil fuels releases CO2 originating from organic matter. This CO2 is therefore depleted in 13C resulting in a decline in the 13C/12C ratio of atmospheric CO254. Accordingly, an atmospheric correction was applied to the tree-ring carbon based on ref. 55. The resulting records contain a small number of missing values (<1%), which we impute by the regularised expectation maximisation (RegEM) method56. All the records were normalised to have zero mean and unit standard deviation.

SPEI reconstruction

We employ a nested principle component regression approach that fits principal component regression models to each set (nest) of tree-ring isotope records and the time series of the gridded SPEI dataset (JA SPEI-02). The Point-by-Point Regression method (PPR)57 has previously been applied to traditional tree-ring networks including in Europe15,58. Using the same well-tested regression reconstruction method allows for direct comparison.

Prior to regression analysis, the PPR method accounts for serial autocorrelation so that the predictand and the predictors are prewhitened prior to calibration. The modelled autocorrelation removed during calibration is added back to the reconstruction to preserve the temporal structure of the instrumental climate data. Single principle component regression models are then fitted for each of the 813 gridpoints and each nest of available proxy records. The pool of available proxy records is restricted to proxy sites within an 800-km radius from the target grid cell and are differentially weighted by the power of 0.5 of its correlation with the predictand variable. Final rescaling of lost variance during calibration relative to the original mean and standard deviation of the climate data is added back to the estimate and ultimately allows a seamless extension with recent climate data until 2018.

Change point and trend analysis

We perform change point analyses to identify periods of significant change using the Wilcoxon rank-sum test. The rank-sum test has been tested suitably for windowing approaches in climate data59,60. Moving window tests of 5 years up to multi-decadal sampling duration are applied to the area-weighted drought reconstruction. To identify changes, tests for the difference in medians are determined by the significance of p values. Due to a large number of test realisations, the 95% confidence level is adjusted for the false discovery rate61. Linear trends were determined by the slope of a linear regression. The significance of trends in the reconstruction was assessed at the p < 0.05 confidence level using a one-sided Student’s t-test, while accounting for lag-1 autocorrelation.