## Introduction

Human-induced climate change is evident and it poses a great concern to society, primarily due to its potential to intensify extreme events around the globe1,2. In the past 2 decades, Europe experienced an increased frequency of droughts3,4 with estimated loss of about EUR 100 billion5. One such devastating event was the drought in summer 2003, which was an exceptionally warm and dry year across most of central and western Europe. Historical reconstructions since 1500 C.E. suggest that it was one of the hottest summers6, and the event was estimated to result in a 30% reduction in gross primary production compared to previous years between 1998–20023. Although, the 2003 drought event was rare and exceptional, even in a multi-centennial time window, its likelihood is expected to increase in the near future7, mainly due to the anthropogenic warming8,9,10,11.

In the summer of 2018, temperature anomaly broke the record again in several locations across Europe, but with distinct spatial patterns. While in summer 2003 the increase in temperature was more concentrated in central and southern Europe (Fig. 1a), summer 2018 was characterised by an anomalous increase in central and north-eastern Europe (Fig. 1b). Unlike the 2003 event—where the temperature anomaly (Supplementary Fig. S1) and the ecosystem carbon and energy fluxes recovered early after the summer12, the extreme event of 2018 persisted to the subsequent year 2019 (Fig. 1c). For all these years, the impact was strongest in the Central European region, where the increase in temperature was accompanied by concurrent significant reduction of summer precipitation (Fig. 1d–f), which led to extreme drought conditions.

The intensity and spatial extent of droughts significantly affects the plant and agricultural productivity13,14, underlying the severity of the drought impact in Central European region, where the focus on agriculture is strong3,7,15,16,17. With the use of remote sensing data-sets18, we find that the concurrent increased temperature with deficit precipitation impaired the condition for vegetation activities (Fig. 1g) in the summer of 2003. We show this in terms of vegetation health index (VHI), which represents the vegetation stress due to the droughts (see methods for detailed description). Similar observations have been made during the summer 2018 as well (Fig. 1h), when several countries suffered agro-economic shocks19. Further, the deprivation of vegetation health persisted and it is noticed even during the summer 2019 (Fig. 1i). In the case of the exceptional 2003 and 2015 events4,20, the vegetation health recovered and returned to its normal condition during the following years. On the contrary, the impact of the 2018 drought on vegetation activities propagated to 2019 and the recovery is still underway, as shown in the time series of VHI (Fig. 1j). Additionally, we note that the vegetation health being categorised in poor condition for at-least 20% of the Central European area in both 2018 and 2019 is unprecedented from the observations in the previous years of twenty-first century. Thus, it is with the utmost urgency that we need to recognise the importance of these persevering consecutive year events, and to develop a holistic framework to model the risk21.

## Results

### 2018–2019 Central European drought from the long-term observational records

The historical reconstruction of composite 254-year long-term climatic database22,23 indicates that although the precipitation anomaly exhibits a drier than average situation during the summer months of 2018 and 2019 across the Central Europe, its intensity is not that high and there are also many other years with similar range of precipitation anomalies (Fig. 1k). On the other hand, 2018–2019 were two out of the three warmest summer periods in the record. To account for this joint effect of precipitation and temperature anomalies, we estimate the drought index based on the standardised precipitation evapotranspiration index (SPEI)24 that considers the atmospheric water supply and demand (see Methods). Our analysis further demonstrates the usefulness of the SPEI estimates as relevant climate predictors for characterising the temporal variability of the summer-time vegetation activities (see Supplementary Fig. S2). While the spatial pattern of summer 2018 SPEI (Fig. 2a) depicts severe drought conditions in the Central European region (SPEI $$\le 0.1$$; see Methods), southern Europe (Balkan countries) experienced wetter than normal conditions (i.e., SPEI $$\ge 0.5$$). Similar to 2018, a severe drought condition (SPEI $$\le 0.1$$) was noticed during the summer of 2019 but the spatial extent of drought was substantially larger compared to the 2018 event (Fig. 2b).

In Central Europe, over 34% of the total land area is extensively used for agricultural purposes19. Our analysis suggests that more than 50% of the Central European region suffered severe drought conditions in the consecutive years of 2018 and 2019. To examine how frequently these consecutive extreme events have occurred in the long-term observational records, we computed and plotted the areal extent of drought ($$A_t$$) with SPEI(t) $$\le 0.1$$ for a given year (t) with the corresponding estimates for the next year ($$A_{t+1}$$) (Fig. 2c). It is evident from the analysis that the 2018–2019 drought is a record breaking event in terms of the consecutive event in the last 254 years, with nearly 50% of the Central European area being classified under the extreme drought conditions. It is also worth mentioning that the 1949–1950 years ranked the second most large-scale consecutive drought years25. Nonetheless, in this case the spatial extent was considerably smaller (around 33%) than that of the 2018–2019 droughts.

The large-scale atmospheric circulation during 2018–2019 was characterized by pronounced positive geopotential height anomalies and anticyclonic circulation pattern at 500 hpa, covering a large area centered over Central Europe and extending to the Northern European region (Supplementary Fig. S3). The complex evolution of these blocking conditions highlights its contribution to the exceptional observed temperature anomalies during 2018–2019. Further, the persistent occurrence of these atmospheric blocking conditions are responsible for the development of large-scale droughts and heat wave, and also triggers soil-moisture temperature feedbacks6,26,27,28, which could further exacerbate and prolong concurrent soil drought and atmospheric aridity29. Literature review suggests that the recent arctic warming is likely to be a main driving factor causing more frequent extreme weather events across the mid-latitudes regions in the Northern Hemisphere30,31,32,33. The major dynamical features for changing the mid-latitude weather due to arctic amplification is the position and structure of the jet stream and planetary wave activity. Jet streams are primarily driven by the difference in temperature between the polar and mid-latitudinal regions. However, the reduced temperature gradient between these two regions has been suggested to lead to a weaker zonal jet with larger meanders32 and that this would cause weather systems to travel eastward rather slowly leading to more persistent weather patterns34. These movement activities are further going to be affected (more persistent) under future warming conditions with increased greenhouse emission35,36. Nevertheless, these theories/mechanisms are still being explored and debatable; and require further rigorous testing37. Moreover, in this study we restrict our focus on analysing (detecting) the exceptional 2018–2019 Central European drought from the long-term observational record point of view, and on the nature of their possible occurrences under warming worlds. Addressing the mechanism (attribution) of the 2018–2019 drought event itself is another line of research, which requires a comprehensive analysis and is beyond the scope of the present study.

### Future occurrences of 2-year droughts under global warming

From our observations, it is clear that, 2018–2019 is an unprecedented 2-year drought event. We now use the ensemble of climate model simulations from the Coupled Model Intercomparison Project phase 5 (CMIP5)38 (Table S1) to understand how the frequency of the 2-year drought event would change in the coming decades, and underpin the role of the anthropogenic warming in exacerbating such drought events (see Methods). In comparison to the simulations based on natural-only forced simulations (HistNat), the occurrence of the 2-year drought event shows a slight increase in the historical simulations (Hist) during the common period of 1850–2005 (Fig. 3a). The differences in the temporal evolution of areas of Central Europe affected by drought among the two sets of simulations have become more apparent during the last 30 years (approx. post 1970; Supplementary Fig. S4)—the period in which there are apparent indications of the role of anthropogenic activities exacerbating global warming39. Climate model simulations based on the Representative Concentration Pathway (RCP) 8.5 scenario further indicates a strong increase in areas under drought towards the end of the twenty-first century (Fig. 3a). Under a moderate scenario of RCP 4.5, the increasing trend persists until the middle of the twenty-first century and stagnates thereafter, while there is apparently no increasing trend in the temporal evolution of area under drought under the more optimistic RCP 2.6 scenario. Based on the climate model simulation results, we find a seven-fold increase in the number of the 2-year drought events, covering at-least one third of Central European domain, in the second half of the century under the RCP 8.5 scenario as compared to the HistNat runs (inset of Fig. 3a). As a result, the corresponding fraction of attributable risk (FAR40; see Methods for more details) under RCP 8.5 is estimated to be nearly one, ascertaining a very strong anthropogenic contribution to exacerbating the occurrence probability of such drought events in the projected future period 2051–2100. Compared to the RCP 8.5 scenario, the number of 2-year droughts events reduces significantly by almost half under the RCP 4.5 scenario and to a very negligible number in RCP 2.6 for the projected period 2051–2100. The corresponding FAR values also reduce to 0.87 and 0.40 for the RCPs 4.5 and 2.6, respectively. These results clearly highlight the diverse role of anthropogenic activities in exacerbating the future occurrence of 2-year drought events; as well the possible benefits of mitigating measures to reduce carbon emissions (encoded as in RCP 2.6/4.5) in lowering the risk of the occurrence of consecutive drought events.

Consistent with the previous studies8,10,41, our analysis also shows that anthropogenic warming will lead to an intensification of European droughts, and to a large extent on the occurrence of 2-year droughts in the future. Such events have substantial implications on many sectors including impacts on agro-phenology, crop water demand and vegetation health activities. Using the long-term historical and projected land use changes based on HYDE database42 (see Methods), we find that drought affected cropland areas across the Central Europe will be nearly doubled (by $$20 \pm 5$$ million ha) under the RCP 8.5 scenario in the second half of the Century in comparison to corresponding historical values (Fig. 3b). This corresponds to the projection of nearly 60% of total cultivated areas being affected by drought in the Central Europe during 2051–2100. Adaptation strategies aiming at the amendment of global warming through the RCP 4.5/2.6 scenarios would significantly reduce the drought prone areas by almost 37%/60%, compared to RCP 8.5. A similar range of benefits in reducing the potential impacts of consecutive year droughts can be expected for areas covered with pastures (Fig. 3c)—which are of high importance for sustaining livestock (i.e., grazing).

## Conclusions

The present study analyses the occurrence of the consecutive droughts over the Central Europe in both historical and projected climate scenarios. The observational record suggests that the ongoing 2018–2019 European drought event is unprecedented in the last 250 years, with substantial implications for vegetation health. Our analysis based on an ensemble of climate model simulations suggests a strong increase in the occurrence of such a rare event, post 2050 under RCP 8.5 scenario. The frequency and the areal extent of these droughts strongly depend on the level of anthropogenic warming scenarios (as encoded in RCPs). Our analysis therefore demonstrates that the occurrences of the consecutive droughts as well their impact on crop and pasture areas can be significantly reduced, if the mitigation strategies leading to amendment of global warming are adopted. One of the major limitations of climate model simulation is its ability to reliably simulate the extreme events and the changes thereof43. Over Central Europe, we notice a general consensus between observations and climate models, especially post-1970, when the anthropogenic influences are apparent (Supplementary Fig. S5). Although, climate models have a relatively good ability to simulate historical past, larger uncertainties may still exist in projections44. Despite this limitation, the climate models are the only available tool to mechanistically understand the occurrence, processes and fate of future extreme events. Our study has mainly focused on detection and the future occurrence of the consecutive drought events. Although we show that under the increased global warming, the observed 2018–2019 droughts are going to increase in the future, an in-depth and separate (careful) analysis is required towards attributing the role of anthropogenic warming in modulating the occurrence of consecutive drought events. Further research is also needed to systematically understand driving mechanisms responsible for such consecutive droughts, whose value to climate adaptation can hardly be overemphasised.

## Methods

### Data

The assessment of consecutive drought characteristics from 1766–2019 is performed over the central Europe using three types of observed gridded meteorologic datasets: Casty et al22 for period 1766–1900, CRU TS dataset23 for period 1901–1949; and E-OBS45 for period 1950–2019. The composite dataset using monthly precipitation and air temperature is analysed at a spatial resolution of $$0.5^\circ \, \times 0.5^\circ$$, similarly as in to previous studies25,46. Furthermore, the E-OBS is used for correcting possible biases in the Casty and CRU data, which is trained on the overlapping period 1950–201525,46. To examine the characteristics of the consecutive droughts in the past and future, we procure the state-of-the-art global climate model simulations from the Coupled Model Intercomparison Project phase 5 (CMIP5)38 (detailed description is provided in Supplementary Table S1). To quantify the effect of human activities in the past (1850–2005), two types of monthly forcings are analysed: (1) natural-only (HistNat), and (2) historical (Hist). While the HistNat contains only the effects of natural forcing (e.g., changes in solar radiation, volcanic eruptions), the Hist considers both, natural and anthropogenic (i.e., greenhouse gas concentrations) effects. To assess possible future climate scenarios, we procure three Representative Concentration Pathway (RCP) scenarios (2.6, 4.5 and 8.5), which are available for 2006–2100. Here, we select 11 climate models based on the consistency in data availability along with the parity in the climate variables (precipitation, temperature and net-radiation) used to estimate SPEI across all the simulations (HistNat, Hist and RCP scenarios). Compared to observations, the climate model simulations were able to capture the overall trend and patterns of atmospheric demand, particularly post-1970–the period when the human influence on the global warming is relatively more apparent39 (Supplementary Fig. S5). In this study, we did not apply bias corrections to the CMIP5 simulations. This is because the employed quantile-based SPEI estimates already account for systematic biases, particularly in the mean and standard deviation, as long as these do not lead to unrealistic P-E dynamics47, which is fairly well captured by climate model simulations (Supplementary Fig. S5).

The vegetation health index (VHI) is one of the important proxies, which is frequently used to evaluate the impacts of drought on vegetation health48,49. This index is applicable for assessing the vegetation stress and to examine the vegetation response to the natural hazards, such in our case, drought49. The VHI for the summer months is obtained from remote sensing data-sets18 at a weekly time step, where it is measured in percentile ranging from 0 to 100. A high value of VHI indicates healthy or unstressed vegetation condition, implying that these areas are not affected by drought conditions (i.e., lack of moisture conditions). The VHI of more than 50% shows above normal and/or healthy vegetation condition. Further, values ranging from 30 to 50% imply vegetation in the region suffering from moderate drought, and the VHI values less then 30% indicate a region experiencing severe drought leading to poor vegetation health conditions. In the present study, we considered VHI values $$\le 30\%$$ as a proxy for poor vegetation health conditions. Subsequently, we inferred the drought-affected vegetation activities as percentage of Central European area (shown in a rectangular box in Fig. 1g) exhibiting VHI $$\le 30\%$$.

### Drought analysis

Recent studies24,50,51,52,53 show the better performance of summer standardised precipitation evapotranspiration index (SPEI)24,54,55 in capturing the drought impacts on hydrological, ecological and agricultural variables than the standard precipitation index (SPI) or the Palmer drought severity index (PDSI). The SPI is not particularly appropriate for our application where both temperature and precipitation are important, since it neither considers the effects of increasing temperature over the recent decades6,56, nor the much larger warming scenario which is expected under future climate change scenarios7. Therefore, the characteristics of drought during summer months (June-August) in Central Europe for both observations and climate model simulations are estimated using SPEI. We used the non-parametric kernel-based approach to estimate the SPEI that can efficiently handle the multi-modality of the sample dataset as compared to other traditional parametric distributions8,57; and it can be represented as:

\begin{aligned} {\mathrm {SPEI}}=F_{t}(x_{t}) \end{aligned}
(1)

where, $$x_{t}$$ denotes the difference between precipitation (P) and potential evapotranspiration ($$E_{p}$$) at a time t. $$F_{t}$$ is the cumulative distribution function estimated using the kernel distribution $$f_{t}(x)$$ of the corresponding time t. $$f_{t}(x)$$ is estimated as

\begin{aligned} f_{t}(x)=\frac{1}{nh}\sum _{t=1}^{n}K \left( \frac{x-x_{t}}{h}\right) \end{aligned}
(2)

where K represents a Gaussian kernel function with a bandwidth h. The h is estimated by the Silverman approach58 for each grid cell separately. The SPEI value using the above-mentioned non-parametric approach varies between 0 and 1, with values below 0.5 indicate drier conditions and above 0.5 the wet conditions. A grid cell in central European region at time t is considered to be in drought when $${\hbox {SPEI}}_{t} \, \le \, \tau$$. Here, $$\tau$$ denotes that the SPEI in the particular grid cell is less than the values occurring $$\tau \, \times 100\%$$ of the time, and the present study considers $$\tau$$ as 0.1 (i.e., 1 in 10-year event or 20% of all dry events)—indicating the occurrence of severe drought event8,59. In case of climate model simulation we also use the non-parametric kernel density estimator, however, we fix the bandwidth with respect to natural-forced historical simulation and use the same for historical; and for all the RCP scenarios considered in the present study. We estimated the yearly development of drought area ($$A_t$$), considering all the cells of the total Central European region that are under drought ($$\hbox {SPEI} \le 0.1$$) for a given year (t). We marked a drought event as a 2-year consecutive event when $$A_t$$ in both years crosses a certain threshold value (e.g., 33.3% or one-third of the Central European region). While estimating the number of consecutive drought events, especially in the RCP 4.5 and 8.5 scenarios, we notice many events with a common (overlapping) drought year. To account for the double counting effect, we counted those events as half which have an overlapping drought year between two (consecutive) events.

Considering the availability of climate variables over long time period (1766–2019), we estimate the monthly potential evapotranspiration ($$E_{p}$$) based from the mean temperature and the approximations for extraterrestrial solar radiation60. Owing to the limitation on the estimation the temperature based $$E_{p}$$61, we check the consistency of this method with an alternative and more physically based $$E_p$$ formulation. In this respect we use two $$E_{p}$$ datasets derived based on the Penman–Monteith method using: (a) the CRU database23 employing the mean, minimum and maximum temperature, vapour pressure, cloudiness and monthly climatology of wind speed available after 1901; and (b) the Princeton Global Forcing (PGF)61 that employs full scale variability of all required meteorological variables (e.g., net radiation, temperature and wind-speed) provided by Sheffield et al.61 for the period 1948–2008. Albeit different underlying meteorological databases being used (CRU23 vs. PGF61), in general, we notice a relatively good agreement among the three $$E_{p}$$ values, especially in capturing the inter-annual variability over the Central European region (see Supplementary Figs. S6 for more details).

We further check the consistency of our results based on the $$E_{p}$$ estimates derived from an energy budget approach following Milley and Dunne62, as given by:

\begin{aligned} E_{p} = 0.8 (R_{n} - G) \end{aligned}
(3)

where $$R_{n}$$ is net radiation at the surface, and G is ground heat flux. Here $$R_{n} - G$$ are estimated using the energy balance as: $$R_{n} - G = L_{v}E + H$$, where $$L_{v}E$$ and H are the latent and sensible heat flux, respectively. Using the climate model simulation outputs, our results show a high correspondence of $$E_{p}$$ between the energy-based approach62 and the Oudin et al.60 for the study domain (see Supplementary Fig. S7a,b for more details). Furthermore, the robustness of our findings on the increased occurrence of the future 2-year consecutive droughts is confirmed, regardless of the employed $$E_{p}$$ methods (see Supplementary Fig. S7c).

### Fraction of attributable risk (FAR)

The FAR has been used by many studies to quantify the anthropogenic influence on the occurrence of recent extreme events and its fate in projected scenarios. The FAR basically addresses the question of what fraction of extremes (in our case 2-year consecutive drought) occurring in Central European region is attributable to anthropogenic influence, and is given by,

\begin{aligned} FAR = 1 - (P_{0}/P_{1}) \end{aligned}
(4)

where $$P_{0}$$ is the probability of exceeding a 2-year consecutive drought without anthropogenic influence (HistNat) and $$P_{1}$$ is with the anthropogenic influence40 (RCP scenarios). The FAR value near to 1 indicates the nearly certain human influence in causing the 2-year consecutive drought.

### Cropland and pasture areas

The impact of droughts on cropland area and pastures are analysed using the dynamics of land use changes of land cover dataset42. This dataset consists of half-degree gridded historical and future fractional land-use patterns and underlying land-use transitions. The historical data uses the HYDE v3.1 historical data set for crop, pasture, and urban area 1500–2005, and the future land cover scenarios 2006–2100 are available for four Integrated Assessment Model (IAM) scenarios which reach different levels of radiative forcing by year 2100:, viz., MESSAGE (RCP 8.5), AIM (RCP 6.0), GCAM/minicam (RCP 4.5) and IMAGE (RCP 2.6). Further, each of these future projections are built by four different historical land-use products, all these are considered in our study.

The cropland cover fraction over the Central Europe started to increase during post 1950, however, a drastic decrease in spatial extent happened after 1990 (Supplementary Fig. S8a). In RCP 4.5 and RCP 8.5 scenarios, the Central Europe will experience a sharp decrease in the overall cropland area. This information is then combined with the fraction of total Central European area which is affected by droughts, as obtained from climate models (Supplementary Fig. S8c). We notice a prominent increasing trend of cropland area affected by drought, especially in the RCP 8.5 scenario. Similar behaviour is projected for the pastures as well (Supplementary Fig. S8b,d). With these observations, we notice a sharp increase in the areal effects both in cropland and pasture by the 2-year consecutive drought in the future, as shown in Fig. 3b. These findings remained same even when we considered a fixed, not time varying, area corresponding to the year 2005 (Supplementary Fig. S8e,f).