This study focuses on meteorological droughts, given its objective of linking drought events to atmospheric circulation. Drought conditions are characterised using the Standardised Precipitation Index (SPI; McKee et al., 1993), as recommended by the World Meteorological Organization (WMO, 2012), and the Standardised Precipitation Evapotranspiration Index (SPEI; Vicente-Serrano et al., 2010). Both indices transform a fitted distribution of a variable accumulated over a given time scale into a standardised normal distribution (μ=0, σ=1). For SPI, the accumulated variable is precipitation, whereas SPEI is based on the climatic water balance, defined as precipitation minus atmospheric evaporative demand (AED).

AED represents the maximum combined evaporation from soil and transpiration from vegetation under given climatic conditions, assuming no soil water limitation. Although AED is often used interchangeably with potential evapotranspiration (PET), from which the index derives its name, this terminology can be ambiguous and is therefore not recommended (Katerji and Rana, 2011; DeJonge et al., 2025). AED is estimated using the Thornthwaite (1948) equation, which does not explicitly parameterise a specific reference surface or vegetation type and requires only temperature and latitude, thereby ensuring methodological consistency across the three datasets used in this study. In western Central Europe, this method agrees closely with the more physically-based Penman–Monteith approach (Penman, 1948). The influence of this approximation on the results is evaluated in Sect. S1, and its implications are discussed in Sect. 5.4.

A 3-month accumulation period (SPEI-3 and SPI-3) is selected as it is sufficiently long to capture a substantial hydrological imbalance, while remaining short enough to be directly linked to dynamic drivers. The 3-month accumulation is calculated using a rolling monthly window, resulting in 12 index values per year. Drought onset occurs when the index falls below -1 (McKee et al., 1993), and drought termination when it rises above this threshold. Drought severity and mean intensity are defined as the sum and mean of the absolute index values during an event, respectively. Drought duration corresponds to the number of consecutive months with index values below -1.

SPI-3 and SPEI-3 are computed using the SPEI R package (Beguería et al., 2014). Following Vicente-Serrano et al. (2010) and Stagge et al. (2015), a Gamma distribution is fitted for SPI-3, and a log-logistic distribution for SPEI-3. The indices are computed relative to the 1947–2008 reference period, corresponding to the common period shared by all post-evaluation reanalyses. The indices are computed at both the grid-cell and regional scales. At the regional scale (48–54° N, 2–10° E), precipitation and AED are aggregated and area-weighted across all land grid cells before the computation of the index to ensure a consistent regional representation, as described in Spinoni et al. (2016). For each month, the drought area involved is also calculated as the proportion of grid cells with index values below -1.

Because meteorological drivers must persist long enough to trigger drought onset (IPCC, 2013; Raposo et al., 2023), the aim is to relate drought events in western Central Europe to low-frequency atmospheric circulation over the Euro-Atlantic sector (30–70° N, 35° W–35° E). The clustering is performed on 500 hPa geopotential height (Z500) anomalies computed relative to the 1947–2008 reference period and focuses on seasonal drought events occurring in February, May, August, and November, corresponding to winter, spring, summer, and autumn, respectively. For each event identified in one of these four months, Z500 anomalies are averaged over that month and the two preceding months (i.e. December–February, March–May, June–August, and September–November, respectively) to match the 3-month accumulation period of SPI-3 or SPEI-3. The k-means clustering algorithm of Hartigan and Wong (1979) is then applied to these composites to identify recurrent circulation patterns associated with drought. Separate analyses are performed for SPI-3 and SPEI-3 drought events.

Clustering is performed in the temporal dimension, with grid cells treated as variables and drought-event Z500 composites as observations. As the k-means clustering algorithm minimises the intra-cluster variance, Euclidean distances are weighted by the cosine of the latitude. To assess clustering stability and reduce sensitivity to centroid initialisation (Philipp et al., 2007), the algorithm was repeated 1000 times for each number of clusters between 2 and 20. The dominant patterns remained stable up to six clusters, indicating limited sensitivity to the initial centroid positions.

Prior to clustering, the number of clusters was fixed to four. Monthly-to-seasonal atmospheric circulation is generally characterised by a few broad-scale recurrent patterns (Huth et al., 2008), unlike daily weather-type classifications that typically require more clusters (e.g. Schwander et al., 2017; Pfister et al., 2025). In addition, only drought-related circulation states are considered here, reducing the expected diversity of patterns. Common clustering diagnostics (Fig. S1) suggested between two and six clusters, providing no unique optimum. Four clusters were ultimately selected based on the spatial separability of the resulting patterns.

To characterise the dominant modes of variability and compare them with the different clusters, the first three empirical orthogonal functions (EOFs) are computed from the complete record of seasonal Z500 over the Euro-Atlantic sector, rather than from drought events only. The EOF analysis is performed on seasonal mean Z500 fields weighted by the square root of the cosine of latitude (Chung and Nigam, 1999).

Temporal trends are detected using the modified Mann–Kendall (MK) test (Mann, 1945; Kendall, 1948) with the variance correction proposed by Hamed and Rao (1998) (hereafter MMKH), implemented through the R package by Patakamuri and O'Brien (2017). The MK test is a non-parametric rank-based method that is widely used for climatological and hydrological time series (Hamed, 2008). Because SPI-3 and SPEI-3 are based on 3-month accumulations, they exhibit serial correlation, violating the independence assumption of the MK test. This effect is accounted for using the variance correction of Hamed and Rao (1998), which considers serial correlation across multiple lags (Khaliq et al., 2009) and has been shown to provide good statistical power (Blain, 2013). Trend magnitude is estimated using Sen's slope (Sen, 1968), a non-parametric estimator defined as the median slope between all pairs of observations.

Regional drought indices (Figs. 2a and S3a) are strongly correlated with the spatial extent of drought (r=0.77–0.81 for SPEI-3 and 0.81–0.84 for SPI-3), indicating that they provide a representative measure of regional drought conditions. The most severe droughts (Fig. 2a), defined as those exceeding the 90th percentile of drought severity over the 1947–2008 reference period for each reanalysis, affected the entire region at their peak (Figs. 2b and S3b), demonstrating high spatial coherence across western Central Europe. Among them, the 1921 drought stands out because of its exceptional duration (14 months in ModE-RA), resulting in a severity exceeding twice the 90th percentile threshold for SPEI-3.

SPI-3 and SPEI-3 identify largely the same drought events (Figs. 2 and S3), with MCC values of 0.76 (ERA5), 0.81 (20CRv3), and 0.74 (ModE-RA) over the common reference period. However, their intensity and severity differ. For example, the well-documented 1976 drought is more intense according to SPI-3, whereas the recent 2018 drought is more intense according to SPEI-3. Since the 1990s, SPEI-3 identifies more droughts, with longer durations than SPI-3 (ERA5, 1990–2023: 33 vs. 25 events, mean duration 2.61 vs. 2.08 months), while mean peak intensity is only slightly higher (1.48 vs. 1.44). Conversely, the most severe droughts are more numerous and generally occur earlier in the record according to SPI-3. In addition to the most severe droughts common to both indices (Figs. 2a and S3a), the 2003 and 2011 droughts appear for SPEI-3 only, whereas the 1853, 1870, 1873, 1893, 1908, 1911, and 1929 droughts appear only with SPI-3.

The recent period 2011–2023 was characterised by frequent droughts, particularly according to SPEI-3 (29.2 drought months per decade in ERA5). Similar periods also occurred during 1857–1874 (31.7 drought months per decade in ModE-RA) and 1947–1960 (29.3–34.3 drought months per decade depending on the reanalysis). In contrast, drought frequency was particularly low during 1912–1920 (1.6 drought months per decade in ModE-RA) and 1977–1988 (6.7–8.3 drought months per decade).

Drought-month count per decade (Fig. 3a) further highlights the pronounced multidecadal variability of drought occurrence. According to SPEI-3, the 2010s experienced the highest drought-month frequency since the beginning of the pre-industrial period, with 35 drought months (29 % of the decade) in ERA5. Western Germany was particularly affected, locally exceeding 40 drought months during the decade. Conversely, the 1910s experienced only five drought months (4 %) according to ModE-RA. These values contrast with the climatological frequency of approximately 20–23 drought months per decade (Fig. 3b), close to the 16 % expected from the standardised drought threshold (SPI/SPEI < -1). The SPI-3 perspective differs substantially (Fig. S4). In particular, the 2010s no longer emerge as an exceptionally dry decade, with only 21 drought months (18 % of the decade), resulting in the largest difference between SPI-3 and SPEI-3.

Year-round SPI-3 and SPEI-3 exhibit contrasting long-term trends (Fig. S5). SPI-3 indicates a long-term wetting over the past 180 years, consistent with increasing 3-month precipitation in all reanalyses (Fig. 4a), with high statistical confidence for ModE-RA and ERA5. In contrast, SPEI-3 exhibits no significant long-term trend, indicating that increasing AED has offset much of the precipitation increase. The strongest increases in both precipitation and AED occur during the most recent period, consistent with accelerated warming. Indeed, their increases per degree of warming are consistent across the three reanalyses (Table S2), despite their different temporal coverage. Warming itself accelerated over time, reaching 0.42 K per decade during 1990–2023 in ERA5.

Seasonal trends reveal a marked contrast between winter and summer. Winter exhibits widespread wetting in both indices, particularly over 1844–2008 (ModE-RA) and 1947–2023 (ERA5), driven by increasing precipitation that outweighs the comparatively small increase in AED (Fig. 4a). Conversely, summer shows the strongest increase in AED, which largely offsets increasing precipitation in ERA5 and 20CRv3, resulting in SPEI-3 drying (low statistical confidence, Fig. S5). Over 1844–2008, ModE-RA indicates summer drying because of both increasing AED and decreasing precipitation. However, this pattern is not reproduced by ERA5 or 20CRv3 over their common period with ModE-RA (Fig. S6), nor by the station observations (Fig. S7).

Spring and autumn generally exhibit intermediate behaviour between the winter and summer extremes (Fig. 4a). Over 1844–2008 and 1918–2015, increasing AED offsets precipitation increases, producing little change in SPEI-3. During 1947–2023, spring is more similar to summer, with SPI-3 wetting but SPEI-3 drying (low statistical confidence), whereas autumn closely resembles winter, with wetting in both indices (high statistical confidence for SPI-3, moderate for SPEI-3). However, station observations do not support the autumn wetting, except at De Bilt, while Uccle shows a statistically significant SPEI-3 drying trend (Fig. S8). The seasonal divergence between SPI-3 and SPEI-3 is greatest in spring over 1844–2008, followed by autumn (Fig. S5), indicating a comparatively strong influence of AED despite smaller absolute increases than in summer.

While the previous results describe aggregated regional trends, local analyses reveal spatial variability (Fig. 4b–c). Agreement between reanalyses is lower locally than regionally (Fig. S8), and correspondence with station observations is less consistent. Nevertheless, several spatial features emerge. Year-round local trends broadly follow the regional signal, except for a significant SPEI-3 wetting along the Dutch coast during 1947–2023. Winter exhibits widespread wetting across the region, in agreement with the station observations. In contrast, spring and summer drying patterns vary spatially and reflect both decreasing precipitation and increasing AED, whereas autumn displays relatively homogeneous wetting despite slight SPEI-3 drying at Uccle.

Four atmospheric circulation patterns have been identified from the clustering of Z500 anomalies during drought events defined by SPEI-3 (Fig. 5a) and SPI-3 (Fig. S9a): a high Z500 anomaly centred over the Baltic region; a high Z500 anomaly centred over the British Isles; a dipole of a high Z500 anomaly over Northern Europe and a low Z500 anomaly over Southern Europe; and a high Z500 anomaly centred over the study area and extending across the European continent. These patterns are hereafter referred to as Baltic High, British Isles High, North–South Dipole, and European High, respectively. For SPEI-3, the Baltic High is less clearly defined in ModE-RA (1844–2008) than in 20CRv3 (1918–2015) and ERA5 (1947–2023). For SPI-3, the centre of the European High is slightly shifted towards the British Isles in ModE-RA and 20CRv3. The circulation patterns identified over the common reference period (1947–2008) exhibit lower spatial agreement between reanalyses than those identified over their respective full periods, likely because fewer drought events are available to robustly characterise the patterns.

The British Isles High (26 %–37 % of SPEI-3 droughts depending on the reanalyses) and the Baltic High (24 %–31 %) are the most frequent circulation patterns associated with drought, whereas the North–South Dipole and the European High occur less frequently. The Baltic High is more often associated with SPEI-3 droughts than SPI-3 droughts, whereas the opposite is true for the British Isles High (Figs. 5a and S9a). Over time, droughts have become increasingly associated with the European High in all reanalyses for both indices (moderate to high statistical confidence). An increase in association with the Baltic High is also observed over 1947–2023. Conversely, associations with the British Isles High and the North–South Dipole declined during 1947–2023 in ERA5 (high statistical confidence) (Figs. 5b and S9b).

The first three dominant modes of variability identified from the EOF analysis correspond to the North Atlantic Oscillation (NAO), the East Atlantic (EA) pattern, and the Scandinavia (SCA) pattern (Barnston and Livezey, 1987; Hernandez and Comas-Bru, 2025; Topál et al., 2025) (Fig. S10). The Baltic High is primarily associated with SCA+ and, to a lesser extent, EA+, whereas the British Isles High is associated with EA-. The North–South Dipole combines SCA+ with NAO-, while the European High is mainly associated with NAO+ over 1844–2008 and 1918–2015 (Figs. 6a and S11a–S12a).

Winter droughts occur predominantly under the North–South Dipole, spring droughts under the European High, whereas the Baltic High and British Isles High occur throughout the year (Figs. 6b and S11b–S12b). Autumn droughts are almost exclusively associated with these latter two patterns. In terms of the dominant modes of variability, winter droughts are characterised by EA- and SCA+, summer and autumn droughts by NAO+, whereas spring shows only weak relationships with the dominant modes (Fig. S13).

Droughts associated with the Baltic High and especially the European High exhibit the largest positive AED anomalies, whereas the North–South Dipole is generally characterised by negative AED anomalies (Figs. 6c and S11c–S12c). Although the magnitude of precipitation anomalies remains substantially larger than that of AED anomalies, AED anomalies associated with droughts increased over time, whereas precipitation anomalies weakened across all periods, particularly over 1947–2023 (Fig. S14).

Drought intensity also varies with the associated circulation pattern (Figs. 6d and S11d–S12d). The North–South Dipole is associated with the weakest droughts, particularly over the central and southern parts of western Central Europe. Baltic High events are associated with stronger droughts over the southeastern part of the region, whereas British Isles High events are associated with stronger droughts over the western part. The European High is associated with the most intense SPEI-3 droughts over 1918–2015 and 1947–2023, especially over western Belgium and northern France. This enhancement is much less pronounced for SPI-3 (not shown), consistent with the larger positive AED anomalies associated with the European High.

Recent severe droughts, particularly the 2018 event, have attracted considerable attention (e.g. Bakke et al., 2020; Zscheischler and Fischer, 2020; Huang et al., 2021). However, the present results show that comparable events have occurred previously in western Central Europe, consistent with earlier studies (e.g. Vicente-Serrano et al., 2021). Notably, the 1921 drought (Bonacina, 1923; Briffa et al., 1994; Bertrand et al., 2021) emerges as the most severe event in western Central Europe over 1844–2023 and has likewise been identified among the most severe European droughts of the past centuries (Hanel et al., 2018; van der Schrier et al., 2021). This conclusion nevertheless depends on the metric considered. Based on precipitation alone (SPI-3), the most severe droughts occur predominantly during the earlier part of the record, in agreement with the long-term increase in precipitation. When AED is included (SPEI-3), recent droughts exhibit an increasing potential to reach comparable severity because of the growing contribution of AED to drought severity (Gebrechorkos et al., 2025). In the future, droughts similar to that of 2018 are expected to become even more severe, partly owing to increasing AED (van der Wiel et al., 2021; Aalbers et al., 2023). Moreover, the exceptional severity of the 1921 drought primarily reflects its persistence (14 months in ModE-RA) rather than its peak or mean intensity. During the growing season, more recent droughts (1947, 2003, and 2018) exceed the 1921 event in severity (Hanel et al., 2018; Hänsel, 2020; van der Schrier et al., 2021). Overall, these findings show that recent droughts are not unprecedented but differ from historical events in both their characteristics and underlying drivers, highlighting the importance of combining multiple drought indices (Lloyd-Hughes, 2014).

Beyond individual events, successive dry periods comparable to 2011–2023 occurred during 1857–1874 and 1947–1960, separated by relatively wet intervals during 1912–1920 and 1977–1988. This pronounced multidecadal variability agrees with previous studies (Briffa et al., 1994; van der Schrier et al., 2006; Hänsel et al., 2019; Haslinger et al., 2019; Bertrand et al., 2021). In particular, the 2010s constitute the decade with the highest drought-month frequency over western Central Europe since the 1850s, extending the findings of Lekarkar et al. (2026) for Belgium over the 1970–2020 period. To investigate the drivers of this multidecadal variability, teleconnection indices were examined, but no substantial relationship was found with the NAO, the El Niño–Southern Oscillation, or the Atlantic Multidecadal Oscillation. Although Atlantic SST variability has been linked to Central European drying (e.g. Haslinger et al., 2019; Tuel and Eltahir, 2021; Haslinger and Mayer, 2023), the associated mechanisms remain uncertain. Likewise, although prolonged periods of reduced precipitation and enhanced AED have been shown to favour multi-year droughts over western Central Europe, the underlying processes remain poorly understood (van der Wiel et al., 2023; van Mourik et al., 2025). Given the greater impacts of multi-year droughts compared with shorter droughts (Chen et al., 2025; Hoover and Smith, 2025; Ruijsch et al., 2025) and their expected increase under warmer climate conditions in western Central Europe (van der Wiel et al., 2023), the successive dry periods identified here provide useful case studies for future research.

Because precipitation predominantly influences droughts in the relatively humid climate of western Central Europe (Raible et al., 2017; Tomas-Burguera et al., 2020), the contribution of AED was expected to be minor. However, the relative contribution of AED during drought events has increased markedly. Since the 1990s, western Central Europe has warmed at 0.42 K per decade (ERA5), exceeding the global mean warming rate (e.g. Dong et al., 2017; Yin et al., 2024; Douville et al., 2025) and making the region particularly sensitive to AED (Huang et al., 2021; Wang et al., 2025). In ERA5, seasonal drought events since 1990 exhibit larger mean AED anomalies (24.3 vs. 7.3 mm) and less negative mean precipitation anomalies (-59.7 vs. -81.2 mm) than those during 1947–1989. These changes are consistent with the concurrent background increases in both precipitation and AED. Thus, although negative precipitation anomalies remain the primary driver of droughts, recent droughts are increasingly characterised by substantial positive AED anomalies as well. This growing contribution is illustrated by the 2010s, which no longer emerge as an exceptionally dry decade when AED is neglected. These findings highlight the importance of systematically accounting for AED when monitoring droughts in western Central Europe, particularly from the 1990s onwards.

The relative influence of precipitation and AED differs when considering background hydroclimatic trends rather than individual drought events. When considering all months, the opposing effects of precipitation and AED on the climatic water balance remain broadly balanced because both variables have increased by comparable magnitudes. Seasonally, however, contrasting responses emerge. The growing divergence between SPI-3 and SPEI-3 reflects the increasing contribution of AED but is not solely controlled by absolute AED trends. Because both indices are standardised, it also depends on the seasonal variability of the climatic water balance. Consequently, over 1844–2008, spring and autumn exhibit the largest increases in the relative contribution of AED, despite smaller absolute AED increases than in summer. AED increased by 14.5 and 13.8 mm in spring and autumn, respectively, compared with 19.5 mm in summer, while the corresponding SPI-3 - SPEI-3 divergences were 0.41 and 0.31, compared with 0.26 in summer. During the more recent periods, however, the much larger increase in AED made summer the season with the largest contribution of AED to SPEI-3 trends. In contrast, winter showed the smallest increase in AED contribution due to the strong energy limitation of winter evapotranspiration, despite significant warming. Nevertheless, winter displays the largest relative increase in AED per degree of warming (33 % K⁻¹ in ERA5), precisely because of its low climatological baseline. Accordingly, winter and autumn became wetter because large precipitation increases exceeded those in AED, whereas summer and, during recent decades, spring experienced drying because large warming-driven increases in AED outweighed the more limited precipitation gains.

The four circulation patterns identified in this study represent distinct hydroclimatic environments over western Central Europe. The Baltic High and the European High are associated with the largest positive AED anomalies, whereas the British Isles High is characterised primarily by negative precipitation anomalies and the North–South Dipole by winter precipitation anomalies combined with negative AED anomalies. These differences reflect the associated flow. The Baltic High induces southeasterly circulation, transporting warm continental air (Serras et al., 2024) that enhances AED, explaining why droughts associated with this pattern are more frequently identified by SPEI-3 than SPI-3. Conversely, the British Isles High favours cooler northerly flow, in line with its association with EA- and the positive relationship between the EA pattern and temperature (Hernandez and Comas-Bru, 2025), resulting in near-zero AED anomalies. The large negative precipitation anomalies associated with this pattern may arise because moist air precipitates over the Scandinavian mountains, leaving western Central Europe in the rain shadow (Pauling et al., 2006). The North–South Dipole occurs predominantly during winter and is associated with SCA+ and NAO-, both linked with cold and dry winter conditions (Bladé et al., 2012; Simpson et al., 2024; Hernandez and Comas-Bru, 2025). This explains its negative AED anomalies, as easterly continental air masses are colder than maritime air masses during winter (Stahl and Demuth, 1999), and consequently why droughts associated with this pattern are generally the weakest. Consistent with the tendency for drought intensity in western Central Europe to increase closest to the largest positive Z500 anomalies (Kingston et al., 2015; Bakke et al., 2023), the European High combines strong negative precipitation anomalies with the largest positive AED anomalies. This agrees with previous studies linking persistent anticyclonic circulation to hot (Stefanon et al., 2012) and dry (Lhotka et al., 2020; Savary et al., 2026) events, as well as their co-occurrence as compound hot–dry events (Ionita et al., 2021), through large-scale subsidence, reduced cloud cover and enhanced incoming shortwave radiation (Kautz et al., 2022).

While the year-round precipitation trends (6 % K⁻¹–8 % K⁻¹, Table S2) are broadly consistent with expected Clausius–Clapeyron thermodynamic scaling (∼ 7 % K⁻¹), the contrasting seasonal trends highlight the role of atmospheric circulation. The evolving association of droughts with these circulation patterns provides a coherent interpretation of the seasonal hydroclimatic evolution, particularly during winter and spring. Winter wetting, characterised by particularly large precipitation increases (12 % K⁻¹–18 % K⁻¹), is consistent with the declining association of droughts with the North–South Dipole, the dominant circulation pattern for winter droughts. Because this pattern is closely linked with NAO-, the observed shift towards more positive winter NAO phases since the mid-twentieth century (Visbeck et al., 2001; Hasanean et al., 2023) provides a plausible explanation for the large increase in winter precipitation, in agreement with the established positive relationship between winter NAO and precipitation over the study area (Simpson et al., 2024). Conversely, spring precipitation increased only weakly over 1947–2023 (3.5 % K⁻¹), well below the Clausius–Clapeyron scaling, whereas AED increased by more than twice as much. This evolution coincides with the increasing association of droughts with the European High, which predominantly characterises spring droughts. Persistent anticyclonic circulation and blocking have repeatedly been identified as the principal drivers of recent spring droughts in Central Europe (Trnka et al., 2009; Ionita et al., 2015; Lhotka et al., 2020; Bakke et al., 2023; Beštáková et al., 2024), supporting the circulation changes identified here.

The increasing association of droughts with the European High, together with the growing influence of AED identified in this study, suggests that compound hot and dry conditions are becoming increasingly characteristic of droughts in western Central Europe. This does not imply, however, that circulation patterns primarily associated with negative precipitation anomalies, such as the British Isles High, will become less relevant under a warmer climate, as negative precipitation anomalies remain a necessary condition for drought development and similar circulation patterns have also been projected to become more frequent (Rousi et al., 2021; Tuel and Eltahir, 2021).

An important strength of this study is the combined use of three independent century-scale reanalyses, each relying on different numerical models, data assimilation systems and assimilated observations. Cross-validation of multiple reanalyses has been recommended to strengthen confidence in analyses of historical climate variability and extreme events (Slivinski, 2018), while reanalyses provide a suitable framework for jointly analysing hydroclimatic variability and atmospheric circulation (Baatz et al., 2021). Evaluation against long weather station observations, together with the comparison over the common reference period, increases confidence in the drought indices. Despite differences between datasets, the principal conclusions regarding drought identification, hydroclimatic trends and atmospheric circulation are consistently reproduced. Remaining discrepancies mainly concern local and seasonal features, reflecting the greater uncertainty of historical precipitation fields and the coarser spatial resolution of reanalyses.

The use of long reanalysis records also highlights the importance of multidecadal variability when interpreting hydroclimatic trends. Sensitivity analyses (not shown) indicate that long-term trends are statistically more robust than trends estimated over recent decades, which are more strongly influenced by the phase of internal climate variability (Hänsel et al., 2019). Consequently, differences between reanalyses primarily reflect their differing temporal coverage rather than contradictory hydroclimatic behaviour. The common reference period provides a practical measure of agreement between datasets, although trends over this relatively short period remain sensitive to its boundaries within the multidecadal variability and should therefore be interpreted with caution. Confidence is highest where reanalyses and station observations converge. Greater caution is warranted where agreement is limited, such as for the decrease in summer precipitation found only in ModE-RA or the long-term autumn wetting found in the reanalyses but not in most station observations.

Using both SPI-3 and SPEI-3 additionally strengthens the interpretation by separating the respective roles of precipitation and AED (Lloyd-Hughes, 2014). This interpretation nevertheless depends on the estimation of AED. The temperature-based Thornthwaite formulation can overestimate warming-driven increases in AED (Seneviratne, 2012; Sheffield et al., 2012), consistent with Sect. S1. However, these differences do not alter the conclusions regarding drought events, long-term trends, or atmospheric circulation. While Penman–Monteith is more physically based, its homogeneous application across the three reanalyses is precluded by the availability and differing estimation of the required meteorological variables (Córdova et al., 2015), and the additional long-term variables introduce their own uncertainties (Trenberth et al., 2014). More generally, the interpretation of long-term AED changes remains complicated by substantial land-use change and rising atmospheric CO₂ concentrations, which are not explicitly represented by either formulation (Villani et al., 2026). Thus, while AED estimation remains a source of uncertainty, its influence on the regional-scale conclusions appears limited, with the evaluation indicating greater uncertainty in long-term precipitation estimates than that arising from the choice of AED formulation.

The identified seasonal circulation patterns are physically meaningful and consistent with previous classifications of Central European drought weather types at daily timescales (e.g. Stahl and Demuth, 1999; Lhotka et al., 2020; Savary et al., 2026). The seasonal framework adopted here accounts for the pronounced seasonality of atmospheric circulation (Comas-Bru and Hernández, 2018), but splits prolonged droughts spanning multiple seasons into distinct events, particularly during the warm season when droughts are generally longer (Banfi et al., 2024). Seasonal mean circulation patterns may also mask important subseasonal variability. In addition, this approach does not directly quantify changes in the frequency or persistence of circulation patterns, although it successfully captures the recent spring drying, winter wetting, and the associated drought characteristics. However, the circulation patterns alone do not fully explain all aspects of the observed hydroclimatic evolution. In particular, the present framework does not explicitly resolve the respective roles of moisture transport, atmospheric convergence, cloud cover, radiative forcing, and land–atmosphere interactions in drought development. Summer drying and some local features indicate that additional processes may also be important. Although changes in atmospheric circulation have been shown to play a major role in recent summer drying (Dunkl et al., 2026), thermodynamic processes also contribute to summer drought development, notably through land–atmosphere feedbacks (Seneviratne et al., 2006; Miralles et al., 2019; Suarez-Gutierrez et al., 2020), reinforced by spring drying through soil–moisture preconditioning (Stegehuis et al., 2021; Douville et al., 2025; Lipfert et al., 2026). Coastal wetting, meanwhile, is likely enhanced by increasing North Sea SSTs (Lenderink et al., 2009; Philip et al., 2020).