Past megadroughts in central Europe were longer, more severe and less warm than modern droughts

Megadroughts are notable manifestations of the American Southwest, but not so much of the European climate. By using long-term hydrological and meteorological observations, as well as paleoclimate reconstructions, here we show that central Europe has experienced much longer and severe droughts during the Spörer Minimum (~AD 1400–1480) and Dalton Minimum (~AD 1770–1840), than the ones observed during the 21st century. These two megadroughts appear to be linked with a cold state of the North Atlantic Ocean and enhanced winter atmospheric blocking activity over the British Isles and western part of Europe, concurrent with reduced solar forcing and explosive volcanism. Moreover, we show that the recent drought events (e.g., 2003, 2015, and 2018), are within the range of natural variability and they are not unprecedented over the last millennium. Central Europe experienced long-lasting droughts during the Spörer and Dalton solar minima around AD 1450 and 1800 that were more severe and extensive than those observed in the 21st century, according to palaeoclimate reconstructions.

D rought is one of the most expensive and damaging natural disasters, which commonly affects large areas and can last for several months to years. Since the beginning of the 21st century, Europe has experienced a series of long-lasting dry and hot summers (2003,2010,2013,2015, and 2018) [1][2][3][4][5] . This type of hydroclimatic extreme can affect all components of the hydrological cycle and it is, usually, associated with significant socio-economic losses 6 . If the deficit in precipitation is combined with high evapotranspiration losses, then it can lead to a deficit in soil moisture and subsequently can manifest itself as a hydrological drought, i.e., deficits in streamflow and groundwater 7 . Prolonged major droughts with severe impacts, such as those recorded in 2003 and 2015 have highlighted Europe's vulnerability to this natural hazard and alerted governments, stakeholders, and operational agencies about the disastrous effects droughts may have on the society and economy, including the need for mitigation measures [8][9][10] . In 2018, the central part of Europe, especially Germany, experienced the warmest April-to-July months since 1880 ( Supplementary Fig. 1a) and in some locations all-time maximum temperatures were recorded. This situation was exacerbated by a rainfall deficit from February to November 2018 ( Supplementary Fig. 1b), when the average precipitation reached, on average, just 58% of the climatological rainfall amount. The analysis shows that this particular longlasting warm and dry year was produced and maintained by a long-lived blocking event 5,11 .
Although the scientific community has attributed the occurrence of these extreme events to anthropogenic climate change [11][12][13] , the instrumental record of precipitation, temperature, and soil moisture is sparse and the satellite record too short, making it difficult to assess whether the 21st-century drying is without precedent. Multicentennial reconstructions of past temperatures for the central part of Europe indicate that the recent summer warming is unprecedented over the last 2500 years 14,15 . Coincident with this increase in temperature, different regions of the European continent have experienced prolonged and severe droughts since the early 2000s 1,2,4,16 , attributed, among other factors, to rising land and ocean temperatures 1,17,18 . For example, the extreme drought events at the European level, in 2003, 2015, and 2018, have been exacerbated by the elevated temperatures 1,12,19,20 , which makes these events unusual during the instrumental period 11,12 . Although these events appear to be unprecedented in terms of temperature contribution, they were also unusual in terms of other characteristics (e.g., magnitude and/or duration). Thus, a comprehensive understanding of the causes and mechanisms of these unusual episodes requires investigations from a long-term perspective.
Here we construct a millennial view for these events, based on long-term hydrological and meteorological data as well as paleoclimate reconstructions of the Palmer Drought Severity Index (PDSI) extracted from the Old World Drought Atlas (OWDA) 21 , for the central part of Europe. In this respect, we make use of a 1000-years tree ring reconstruction of summer drought at European level 21 , to evaluate whether the recent drying was unprecedented in its joint duration and severity over the Common era. The focus here is on central Europe (black square in Fig. 1a), given that in 2015 and 2018 this was the most affected region by the extreme drying and that a rich database of climate reconstructions is available for an in-depth evaluation of the drought occurrence over the last millennium. We also make use of sea surface temperature (SST), sea surface salinity, and large-scale atmospheric circulation reconstructions and paleo reanalysis to investigate the driving mechanism behind the dry periods over the last millennium.

Results and discussion
The reconstructed PDSI index, based on the OWDA, for the analyzed region accounts for 52 % (r = 0.71, p ≪ 0.001) of the variability in the observed self-calibrated Palmer Drought Severity Index (scPDSI index) over the common period (1901-2012) (Fig. 2a). For the common period , the driest years are 1921 and 1976, respectively (Figs. 1b and 2a). The years 1102, 1503, 1865, and 1921 are recorded as the driest ones over the last millennium, based on the OWDA data ( Fig. 2a), while summers 1949, 1976, and 1990 are recorded as the driest ones over the observational period, based on the observed scPDSI (Fig. 1b). The 1921 drought had a higher amplitude over the north-western part of Germany ( Supplementary Fig. 2), while the 1976 drought was recorded over the whole analyzed region. These aspects are also captured by the low-flow situation on the Rhine, Elbe, and Weser rivers ( Supplementary Fig. 3). The year 1921 ranks also as the driest in the Rhine and Weser catchment areas (Supplementary Fig. 3b and 3c) (both situated in the western part of Germany). The years 2003, 2015, and 2018 do not rank among the top driest years either in the reconstructed PDSI or in the observed records. From 1810 onwards, the driest years (1865, 1893, 1921, 1947, 1976, and 2003), as captured by the reconstructed PDSI index, correspond also to extremely dry years, in terms of low flow periods, in the observed streamflow data for the most important rivers in central Europe (the Rhine, Elbe, and Weser, Supplementary Fig. 3). This is an indication that the reconstructed PDSI index can properly capture the occurrence of drought events over the analyzed period, and can be used to place the 21st-century droughts into a long-term context. From the perspective of the last~1000 years, the 20th century and the beginning of the 21st century are characterized by reduced multidecadal variability in the occurrence of dry and wet periods, mainly when compared with the previous centuries (Fig. 2a). The driest years, over the last millennium, in the central part of Europe are 1102, 1419, 1503, 1504, 1858, 1865, and 1921, respectively ( Fig. 2a and Supplementary Fig. 4).
Two megadroughts in central Europe. Throughout the last millennium, two distinct and long-lasting dry periods, so-called megadroughts, are observed: one over the~1400-1480 period, and the other over the~1770-1840's time interval (Fig. 2a, b). The mid-15 th century megadrought largely was synchronous with the Spörer Minimum (1420-1550), a period of low solar activity with the strongest reduction in incoming total solar irradiance (TSI) 22 , which was punctuated also with several large volcanic eruptions 23 . In Western Europe, the early Spörer Minimum was coincident with a series of extremely cold and long-lasting winters [24][25][26][27] , which had a dramatic effect on the productivity of terrestrial ecosystems in the subsequent growing seasons, manifested also as devastating losses in agricultural production. Based on documentary evidence 28 , here we show that the 1400-1480 megadrought event was characterized by dry summers (the driest decade over the last millennium was recorded between 1471 and 1480, Fig. 3d), and colder than average winters (Fig. 4b) and springs (Fig. 4c). A tree-ring based reconstruction of spring (AMJ) precipitation ( Supplementary Fig. 5d) and summer (JJA) temperature ( Supplementary Fig. 5e), over a larger region covering the western and central parts of Europe, indicates that this period was characterized by a decline in the mid-spring precipitation and cold summers, especially over the 1440-1480 29 period.
Similarly, the 1800's megadrought corresponds to the Dalton Minimum (1790-1830), also a period of low solar activity and several volcanic eruptions, which coincided with a period of lower-than-average global temperatures. During this period, there was a variation in temperature of about −1°C in the central part of Europe 28,30 . The 1770-1840's megadrought event is associated with one of the driest periods over the last millennium (Fig. 3). Winter, spring, and summer seasons were all characterized by negative precipitation anomalies over more than five consecutive decades ( Fig. 3b-d) and colder than normal winters and springs (Fig. 4c, c). This extremely dry period was accompanied by cold and harsh long-lasting winters in the Baltic Sea 31 . The second megadrought event (~1770-1840) is clearly captured also by a new reconstruction of the Standardized Precipitation Index over Germany, for the last 500 years ( Supplementary Fig. 5b) 32 . This event was also observed in Poland, where the 1770-1840 period was characterized by long-lasting dry events and precipitation deficit ( Supplementary Fig. 5c). In other reconstructions 29,33,34 , spanning the last millennium, the driest period was recorded between 1770 and 1820 (Supplementary Figs. 5, 6, and 7), while over the 1820-1840 period altering of dry/wet years can be observed. The differences between the different data sets employed at the end of the second megadrought (~1820-1840), might be due to the different types of proxy data used to derive the reconstructions and their uncertainties.
To emphasize the differences between the two megadroughts and the present-day drought variability, in Fig. 2c we show the statistical distribution of reconstructed PDSI (OWDA) index for three distinct periods, corresponding to the two megadrought events (1400-1480-red line; 1770-1840-orange line, and the 1901-2012 period gray line), and of the observed scPDSI index over the 1901-2018 period (blue line). The Gaussian distribution fitted over the four periods shows that there is a significant difference in the distribution of the two mega-dry periods when compared to the 20th and 21st century drought events, indicating that the megadroughts are associated with anomalous climate regimes. From Fig. 2c we can infer that the 2003, 2015, and 2018 droughts are not unprecedented over the last millennium.
Links between solar forcing and the megadroughts. Numerical integrations indicate that high/low solar irradiance levels are generating weak/strong AMOC states, a few decades later 35,36 . Consistent with this, the TSI maxima just before 1400, 1600, and 1800 are followed by weak Atlantic Meridional Overturning Circulation (AMOC) episodes, a few decades later (Fig. 5d) 37 . Due to the fact that TSI evolves from a positive to a negative also in several decades, the weak AMOC state is quasi-synchronous with the TSI minima ( Fig. 5d-f). Model simulations show also that a negative Atlantic Multidecadal Oscillation (AMO)/ weak AMOC state induces an in-phase baroclinic atmospheric response consisting of a high-pressure system extending from eastern North Atlantic towards northern and central Europe 38,39 . A similar atmospheric structure, e.g., an anticyclonic circulation over the British Isles and the western part of Europe, associated with periods of low solar activity, is derived based on proxy and paleo reanalysis data (Fig. 6a, b). Therefore, through this causal chain, TSI minima are associated with such a high-pressure system, consistent with numerical integrations 40,41 . The atmospheric center of positive Z500 anomalies (Fig. 6) largely suppresses ascending motions, reduces water vapor condensation and precipitation formation, leading to drought conditions below this atmospheric system. This effect is amplified by the inherent persistence of this typical blocking configuration. Consistent with this, dry summers are associated with a seasonally persistent high-pressure system (from winter to summer) centered over the British Isles and the western part of Europe ( Supplementary  Fig. 8). The association between cold North Atlantic conditions, the high-pressure system and the dry summers in central and northern parts of Europe manifests also over the observational period, when extended dry periods over the analyzed region (e.g., 1971-1976) (Fig. 7a) occurred in combination with a cold North Atlantic basin (Fig. 7c) and enhanced atmospheric blocking over the central part of Europe (Fig. 7d). The 1971-1976 time interval was characterized also by an abrupt AMOC weakening 42 .
The two megadroughts over the central part of Europe and their relationship with the prevailing SSTs in the North Atlantic basin are visible also in two paleo reanalyzes of the last millennium ( Supplementary Figs. 7, 9, and 10) 33,34 , although the amplitude of the megadroughts in the paleo reanalysis data is smaller compared to the amplitude of these two events in the Old World Drought Atlas (OWDA) reconstruction ( Supplementary  Fig. 7). These links between cold North Atlantic conditions and European scale drought are supported also by previous studies based on observational data [42][43][44] . Similarly, it was shown that during the negative phase of AMO (cold North Atlantic basin) Germany and the southern part of the Scandinavian Peninsula is affected by dryness, while a warm North Atlantic basin (positive AMO phase) is associated with wetness over these regions 45 . The canonical understanding of oceanic influences on European droughts suggests that, on multidecadal time scales, summer drought variability over the central and western part of Europe is strongly associated with the variability of the North Atlantic SSTs in the previous winter 42 . In general, dry (wet) summers over the central and western parts of Europe are associated with cold (warm) SSTs in the North Atlantic basin in the previous winter ( Fig. 7 and Supplementary Fig. 11).
Here we used independent reanalysis data sets and proxy records to show that, concurrent with low solar activity, cold North Atlantic conditions linked to weak AMOC states are associated with a high-pressure system located over central and northwestern Europe and enhanced blocking activity (Fig. 7), resulting in long-lasting dry periods in the mid-15 th century and the beginning of the 18th century. Unlike the megadroughts throughout the last millennium, the worst-case droughts of the 21st century (e.g., 2003, 2015, or 2018) are also amplified by the extremely warm summers, in association with the global warming, which will likely exacerbate drought over the European continent 46 . The fact that for other solar minima periods (e.g., Maunder Minimum,~1655-1715) we do not see dry periods over the western part of Europe, indicates that the 15th and 18th-century megadroughts are the response to both internal and forced climate variability. During the Maunder Minimum, paleoclimate reconstructions indicate that the North Atlantic was warmer compared to the Spörer and Dalton minimum time intervals (Fig. 5) 47 and that AMO was in a positive phase 48 . This period was associated with precipitation excedent over our analyzed region (Fig. 3). Thus, the two western European megadroughts might have been the result of an interplay between solar extremes, cold North Atlantic surface waters, enhanced blocking activity and explosive volcanism.   6 Large-scale climate drivers of dry events. a The low composite map between the reconstructed 21 PDSI index (PDSI < − 0.75 std. dev.) and winter reconstructed Geopotential Height at 500 mb (Z500) 53 and (b) the low composite map between the reconstructed TSI 63 (TSI <−0.75 std. dev.) and winter reconstructed Geopotential Height at 500 mb (Z500) 53 . The hatching highlights significant values at a confidence level of 95 %. Analyzed period: 1500-1999. Units: Z500 (m).

Conclusions
By using different independent data sets (e.g., observations, paleo reanalysis, documentary evidence, and proxy records) in this study we provide a comprehensive assessment of past megadroughts in central Europe and their underlying drivers. Moreover, we have shown that the recent droughts (e.g., 2003, 2015, and 2018, among others) are within the historical variability and they are not unprecedented over the last millennium. Future climate projections indicate that Europe will face substantial drying, even for the least aggressive pathways scenarios (SSP126 and SSP245) 49 . Although the greenhouse gases and the associate global warming signal will substantially contribute to future drought risk 49 , our study indicates that future drought variations will also be strongly influenced by natural variations. A potential decrease of TSI in the next decades could result in a higher frequency of drought events in central Europe, which could add to the drying induced by anthropogenic forcing. The potential manifestation of record extreme droughts represents a possible scenario for the future and it would represent an enormous challenge for the governments and society. Thus, determining future drought risk of the European droughts requires further work on how the combined effect of natural and anthropogenic factors will shape the drought magnitude and frequency.

Methods
Observational, reanalysis and proxy data. Observed precipitation, temperature, and scPDSI were obtained or calculated from the Climate Research Unit's TS 4.03 data product 50 . These fields have a 0.5°× 0.5°resolution. The streamflow data was provided by the German Hydrological Institute (www.bfg.de). As observed SST fields, we have used the HadISST data set, with a 1°× 1°resolution 51 . For the largescale atmospheric circulation, we have used geopotential height, zonal and meridional wind at 500 mb from the 20th Century Reanalysis data 52 , as well as reconstructed geopotential height at 500 mb (Z500) over the Eastern North Atlantic and Europe, back to 1500 53 . The seasonal reconstructed precipitation ( Supplementary Fig. 6), over the European region, is based on two distinct data sources 54,55 .
Drought reconstruction. From the Old World Drought Atlas (OWDA) 21 we extracted all grid points in the range 3°E-20°E and 45°N-56°N, for the 1000-2012 period and calculated a PDSI index, by averaging all the grid points from the aforementioned region. To make valid seasonal comparisons to the current drought conditions based on observational records (CRU TS4) 50 , we scaled the OWDA time series by the mean of the observations and verified their standard deviations. These two-time series are highly similar (r = 0.71, p ≪ 0.001 for the 1901-2012 period and r = 0.79, p ≪ 0.001 over the 1901-1978 period). We estimated the uncertainty in the scaled reconstruction by calculating the root-meansquared error from the residual of the fit between the instrumental and original paleoclimate data. The results are shown in Fig. 1. One of the reasons for which we chose the current regional setup (e.g., central part of Europe) is also due to the fact that the OWDA has the proven capacity to reconstruct drought variability at multidecadal time scales, especially over this region. The highest tree ring network, used in the OWDA development is located over Germany (Fig. S1 in Cook et al., 2015 21 paper). The OWDA shows the strong agreement with the seasonal precipitation reconstruction over Europe, for the last 500 years 55 , especially in parts of central Europe (e.g., Germany, also in the pre-instrumental period).
Historical data. The data in Figs. 3 and 4 originate from different sources that cover the German territory and the neighboring countries (see Fig. 1 in Glaser and Riemann (2009) 56 ). The seasonal decennial index represents the difference for each decade between too warm and too cold, and too wet and too dry years. The criterion for exceeding or falling below 'normal' events is the 0.75 standard deviation from the mean, relative to the period 1951-1980. A more detailed description of how the decennial index is computed is given in Glaser (2013) 28 . A wide range of historical data has been used to derive the seasonal decennial indices for temperature and precipitation, like annual town chronicles of German cities, family chronicles, diaries, newspapers, administrative records, harvest records, as well as proxy data (e.g., tree rings). Most of the data can be accessed at www. tambora.org, where a detailed description for each source type is given. To calibrate the precipitation and temperature indices extracted from the documentary data, instrumental data was used 28 . This index represents one of the longest and welldated reconstruction in the central part of Europe regarding the seasonal variability of wet/dry and cold/warm regimes, over the last 1000 years.
Last millennium reanalysis. To test the potential linkages between the drought variability and the oceanic conditions over the last millennium, in this study we employ two last millennium reanalysis data: PHYDA data set 33 and Last Millennium Reanalysis (LMR) fields 34 . PHYDA reanalysis data is based on assimilation, a methodology that combines proxy information with global climate simulations 33 . PHYDA comprises information from a network of 2970 annually resolved proxies, together with the last millennium ensemble simulations based on the Community Earth System Model 57 . PHYDA outputs (e.g., PDSI and 2-m surface air temperature) span the years 0-2000 CE and have a spatial resolution of 2.5°lat ×~1.9°lon. This dataset has been developed with the specific aim of studying hydroclimate variability over the Common Era. PHYDA has been validated against observational data 33 and has been successfully employed to study oceanic and radiative forcing of the megadroughts in the American Southwest 58 . The Last Millennium Reanalysis (LMR) 34 uses an ensemble methodology to assimilate paleoclimate data to produce annually resolved climate field reconstructions of the Common Era (PDSI, SST, and geopotential height). LMR combines, through data assimilation, proxy data, and model climate simulations. The data assimilation is performed using a variant of the ensemble Kalman Filter. In this study, we use version 2.0 59 , which includes the PAGES2K database 60,61 as a source for the proxy records and the CCSM4 last millennium simulations 62 . From the PAGES2K database, the only series which are annually resolved have been considered. LMR makes it possible to validate dynamical hypotheses for the causes of megadroughts throughout the simultaneous reconstruction of drought indices, ocean variables, and large-scale atmospheric circulation fields. Because the output of the LMR has an annual resolution, in this study we make use only of the global SST field, to study the relationship of the summer drought over the analyzed region in relationship with the global SST over the last millennium.
Composite analysis. To identify the physical mechanism responsible for the connection between the drought occurrence and the large-scale atmospheric circulation, we constructed the composite maps between the normalized time series of PDSI (TSI) for the years when the values of the index were lower than 0.75 std. dev ( Fig. 6 and Supplementary Fig. 8). This threshold was chosen as a compromise between the strength of the climate anomalies associated with large-scale anomalies and the number of maps that satisfy this criterion. Further analysis has shown that the results are not sensitive to the exact threshold value used for our composite analysis (not shown). We have computed composite maps, instead of correlation maps, because the former reflect the nonlinearities included in the analyzed data. The significances of the composite maps are computed based on a standard t-test (confidence level 95%).

Data availability
No datasets were generated during this study. All data used in this study are publicly available from the sources cited in "Data and Methods".

Code availability
All code necessary to perform the reported analyses can be obtained from the corresponding author.