2.1 Drought Indices

The severity of droughts was assessed for seven long-term stations located in the Czech Republic, their elevations ranging roughly from 200 to 450 m above sea level (a.s.l.) (Figure 1a). The sites represent weather stations with the longest daily precipitation and temperature records available within the Czech Hydrometeorological Institute database. The study applies four drought indices (Table 1).

PDSI (Palmer, 1965) was one of the first historical attempts to identify droughts using more than just precipitation data. Due to the fact that PDSI was originally calibrated with empirical constants, soil properties, and climate characteristics derived from nine stations in Kansas and Iowa (USA), its self-calibrated version (Wells et al., 2004) with a modified snow-melting module (Trnka et al., 2010) was used. In general, PDSI is based on the water supply-and-demand concept and incorporates precipitation, temperature, and water-holding capacity of soils (177.8 mm [7 in.] of precipitation was used identically for all stations). First, anomalies of water balance were calculated based on a difference between precipitation and potential evapotranspiration. Potential evapotranspiration was estimated through Thornthwaite method that is based on temperature data and latitude of a station. From the series of water balance and considering the calibration period (1912–2015), the Palmer Z Index (PZIN) was calculated. Finally, PDSI is then computed directly from PZIN and shows a time series of cumulative departures (with respect to previous conditions) in atmospheric moisture supply and demand – the complete algorithm is presented in Jacobi et al. (2013). Because PZIN is calculated directly from the water balance anomalies (not considering previous conditions), it responds to short-term episodes better than does PDSI and enables identifying rapidly developing drought conditions (WMO and GWP, 2016).

Another drought index applied is the SPEI (Vicente-Serrano et al., 2010). Analogously to PDSI, it is calculated based on anomalies in water balance (precipitation minus potential evapotranspiration); however, SPEI does not consider water-holding capacity of soils. Anomalies of water balance were then aggregated at 4 and 52 weeks timescales (SPEI-4 and SPEI-52) and modeled using log-logistic statistical distribution. Full documentation of SPEI is presented in Vicente-Serrano et al. (2010). In addition, a procedure that classifies precipitation as rain or snow was applied (Trnka et al., 2010). The SPEI-4 index characterizes rather short-term drought, while SPEI-52 is more suitable for analyzing long-term events.

All drought indices were calculated from temperature and precipitation data as a meteorological input, because other drought-related meteorological variables (e.g., solar radiation, wind speed, and humidity) are not available in these historical records. The indices are available in weekly temporal resolution (52 values per year) from the start of the individual stations' observation to 2018. The longest records are for Prague-Klementinum and Tabor (1876–2018, 143 years) while the shortest series at Klatovy starts in 1922 (97 years).

The individual drought indices respond differently to changes in input variables. Distributions of weekly changes in the drought indices (hereafter referred to as tendencies) are shown in Figure 2 (example for Prague, the results are analogous for the other stations). The tendencies were calculated separately for the vegetation period (VEG, weeks 14–39, roughly April–September) and the cold period (COLD, weeks 40–13, approximately October–March). There were significant differences between VEG and COLD in the shapes of their distributions (according to the Kolmogorov-Smirnov test at the 5% level), except for SPEI-4. The PDSI, PZIN, and SPEI-52 distributions had lower kurtosis in VEG, indicating higher frequency of extreme values compared to COLD. In addition, PDSI was positively skewed in both seasons, which suggests, for example, that positive tendencies larger than 0.25 are more abundant than negative tendencies below −0.25 (Figure 2). This is related to inclusion of the soil water-holding capacity in PDSI. The maximum soil water-holding capacity can be reached in a relatively short time during heavy rains; however, its depletion is a rather long-term process (Palmer, 1965). As expected, the short-term drought indices (PZIN and SPEI-4) exhibited larger values of weekly tendencies compared to the long-term ones, and thus, they are more suitable for linking drought to changes in atmospheric circulation.

2.2 Circulation Indices and Types

Atmospheric circulation over central Europe was characterized by circulation types derived from circulation indices. The indices (flow strength, direction, and vorticity; Jenkinson & Collison, 1977; Blenkinsop et al., 2009) were calculated at the daily timescale using sea level pressure at 16 points regularly distributed across Europe, centered at 50°N and 15°E (Figure 1b). The sea level pressure data were taken from the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis (Kalnay et al., 1996), which is available from 1948 onward. The first circulation index is a flow strength (STR), which is a vector sum of western (w) and southern flow (s) components:

where P(1–16) indicate sea level pressure value for a given point in hectopascals. The second circulation index, flow direction (DIR), is calculated using a multivalued inverse tangent function and is given by the formula:

The third circulation index is flow vorticity (VORT), which is the sum of its zonal (zw) and meridional (zs) components and reflects the rotation of an air mass. It indicates anticyclonic (VORT < 0) or cyclonic (VORT > 0) weather conditions and is calculated using the formulas:

In the next step, 27 circulation types based on Lamb (1972) catalogue were derived from the STR, DIR, and VORT indices. This approach was originally introduced by Jenkinson and Collison (1977) and later modified by Plavcová and Kyselý (2011) for central Europe. If STR and VORT values were lower than 4, the pressure field lacked distinctive features and that day was left unclassified (U). When the absolute value of VORT is at least five times larger than STR, strongly anticyclonic (A, if VORT < 0) or strongly cyclonic (C, if VORT > 0) type was assigned. If STR was larger than the absolute value of VORT, the day was classified into one of the eight directional circulation types (N, NE, E, SE, S, SW, W, and NW). The remaining days were assigned to hybrid circulation types based on their direction and anticyclonic or cyclonic vorticity (e.g., ASW and CW). This classification was found suitable for linking atmospheric circulation to near surface temperature (Plavcová & Kyselý, 2011), precipitation (Plavcová et al., 2014), and heat waves (Lhotka, Kyselý, & Plavcová, 2018) in central Europe. Mean sea level pressure patterns related to individual circulation types are shown in Figures S1–S3 in the supporting information.

Frequencies of the circulation types in VEG and COLD are presented in Figure 3. Overall, strongly cyclonic and hybrid cyclonic circulation types were less abundant compared to their anticyclonic analogues and directional circulation types during both seasons. For VEG, the largest frequency was found for the A type (7.2%), and AW, ANW, W, and U all had frequencies larger than 5%. In COLD, by contrast, advection from the western quadrant dominates, and ASW, AW, SW, and W circulation types accounted for more than one third of days in the 1948–2018 period. The pressure field in COLD is generally more distinctive, and therefore the U circulation type appeared far less frequently in COLD compared to VEG (1.1% vs. 5.3%; Figure 3).

2.3 Linking Drought Indices to Atmospheric Circulation

Relationships between droughts and atmospheric circulation were studied through weekly tendencies of short-term drought indices (PZIN and SPEI-4), which have larger signal-to-noise ratios due to the larger changes between individual weeks (Figure 2). The indices were averaged across all seven stations and differences between two consecutive values were calculated. A negative tendency suggests drier conditions compared to the previous week and vice versa. Using actual drought values would not be straightforward due to a relatively long soil moisture memory (Delworth & Manabe, 1988). The drought tendencies in weekly resolution, however, cannot simply be linked to circulation types, which change on shorter timescales (Table S1). In order to overcome this issue, as many as seven daily circulation types were assigned to one weekly drought tendency value. When a single circulation type was classified during all days of the given week (which happens rather rarely), it received a weight of 7. Analogously, if two or more different circulation types occurred in the given week, they were given weights between 1 and 6 according to their duration in days within that week. For each circulation type a weighted average of tendencies was calculated for COLD and VEG in order to analyze its conduciveness to drought.

3.1 Temporal Variability of Drought Events

Using PDSI calculated for seven stations across the Czech Republic, long-term temporal variability of drought events was analyzed (Figure 4). The most distinctive feature recorded at all stations was the presence of extreme droughts (PDSI value lower than −4) from the 2000s onward. PDSI dropped below −6 in 2018 at stations Brno and Prague, demonstrating drought conditions unprecedented since the late 19th century (Figure 4). In Prague, extreme drought was observed from summer 2015 onward (for more than 3 years), creating the longest consecutive extreme drought event. Extreme drought episodes were recorded also prior to the 2000s, but these were not spatially coherent, for example, around 1945 (Klatovy) and 1985 (Ceske Budejovice).

All stations had negative PDSI trends (considering their whole periods of observation) ranging from −0.13 to −0.38 per decade, and similar trend values were found in the overlapping 1922–2018 period (Table 2). All trends were significant at the 5% level according to the modified Mann-Kendall trend test for autocorrelated data (Yue & Wang, 2004). Larger differences in the PDSI trends were observed during 1948–2018, a period covered by the NCEP/NCAR reanalysis that provides sea level pressure data for assessing atmospheric circulation. The most pronounced PDSI trends were found in Brno and Prague (−0.54 per decade), while the trend was close to 0 in Ceske Budejovice due to severe droughts in the 1950s and 1980s (Figure 4). The most recent 1979–2018 period (its onset corresponding with the beginning of satellite measurements) was characterized by the largest negative PDSI trends (especially in Brno, Prague, and Caslav; Table 2). Analogously to the earlier period, an indistinct statistically insignificant trend was observed in Ceske Budejovice, mainly due to the 1980s extreme drought.

In the next step, we analyzed drought episodes in more detail in VEG and COLD. The time series were averaged for all seven stations, and therefore the limited overlapping 1922–2018 period was studied. Considering PDSI, the 2015–2018 period had by far the lowest values in both seasons (with a minimum of −4.5 in VEG during 2018 and the same value was found in COLD during 2015/2016, Figure 5). This period fell outside the confidence interval defined as the 0 ± 2 × standard deviation (SD) of mean seasonal PDSI for 1922–1999. PDSI values below this interval were recorded also in 1991, 2007, and 2008 for VEG and during 1983/1984 and 2008/2009 for COLD. Confidence intervals were calculated analogously for the other drought indices.

A similar pattern was found considering another long-term index, SPEI-52. The last 4 years of observation were extremely dry especially in VEG, but the record-breaking SPEI-52 value (−2.3) was found for COLD in 2015/2016. During VEG, the record-breaking SPEI-52 (−1.8) occurred in 2007 but the 2018 season ranked as the second driest. In general, the PDSI and SPEI-52 indices showed similar temporal variability patterns for VEG and COLD, mainly due to their long-term nature—the respective seasonal values for VEG were relatively strongly correlated to those for the previous COLD season (the Pearson correlation coefficient [r] was 0.85 for PDSI and 0.74 for SPEI-52). In addition, a strong correlation between seasonal PDSI and SPEI-52 values (Figure S4) suggests a good agreement between these two indices in their ability to capture long-term drought conditions.

Differences between VEG and COLD were more pronounced for short-term indices, and seasonal values of PZIN for VEG were only weakly related to those for the previous COLD season, as expected (r = 0.37). Moreover, seasonal SPEI-4 values for VEG were nearly independent of those for COLD (r = 0.16). During VEG, short-term indices revealed rapidly developing droughts in 1947, 2015, and 2018. These three seasons were linked to the three lowest values of both PZIN and SPEI-4. The driest COLD was recorded in 1953/1954, with the PZIN value of −0.7 and SPEI-4 value of −1.0. Only two other years fell outside the confidence interval in COLD as well (2006/2007 and 2013/2014), and there was good agreement between PZIN and SPEI-4, as shown in Figure S4. The long-term (PDSI and SPEI-52) indices showed significant negative trends for both seasons (using the Mann-Kendall trend test at the 5% level), which was associated with the extremely dry recent 2015–2018 period (Figure 5). Considering short-term indices, however, statistical significance was determined for PZIN in VEG only.

3.2 Links to Atmospheric Circulation

Relationships between drought indices and circulation types were analyzed for the 1948–2018 period, for which sea level pressure data from the NCEP/NCAR reanalysis were available. These circulation types were linked to drought tendencies (section 2.3) using short-term PZIN and SPEI-4 indices. Figure 6 shows mean drought tendency for each circulation type in COLD and VEG, where negative tendencies suggest increase of dry conditions and positive tendencies correspond to a wetter regime. The anticyclonic circulation types had negative mean drought tendencies (down to −0.29 considering PZIN and −0.15 for SPEI-4). By contrast, all cyclonic circulation types were related to positive mean drought tendencies and their magnitudes were larger compared to those under anticyclonic circulation types (up to 0.55 considering PZIN and 0.31 for SPEI-4). The larger absolute values of mean drought tendencies under cyclonic circulation types were compensated by lower frequencies of these types (Figure 3). Positive drought tendencies were found also under most directional circulation types except for the easterly (E), southeasterly (SE), and southerly (S) types (Figure 6). Although a relatively large difference between mean drought tendencies in COLD and VEG appeared for some circulation types, both values always remained completely within the positive or negative sector (Figure 6). The only exception was the CS circulation type (with a low frequency of occurrence) linked to SPEI-4, which had positive mean drought tendency in COLD and slightly negative in VEG.

Based on these results, 12 circulation types with negative mean drought tendencies (averaged between PZIN and SPEI-4) were defined. Those were all anticyclonic types (A, AN, ANE, AE, ASE, AS, ASW, AW, and ANW) and directional types from the southeast quadrant (E, SE, and S; hereafter referred altogether as dry circulation types). The remaining 15 circulation types, that is, all cyclonic types (C, CN, CNE, CE, CSE, CS, CSW, CW, and CNW), directional types with northern and/or western component (NE, N, NW, W, and SW), and undefined type (U), had positive mean drought tendencies and were labeled as wet types.

The annual frequency of dry circulation types during VEG had a statistically significant (at the 1% level, according to the Mann-Kendall test) positive trend in the 1948–2018 period (roughly 2% per decade), with the largest value (68%) found in 2015. During the last 30 years, dry circulation types occurred more frequently than wet ones in 23 out of 30 VEG seasons, which is a completely opposite pattern compared to that observed in the 1950s and 1960s (Figure 7a). The droughts that fell outside the 2 × SD confidence intervals (section 3.1) occurred in those seasons in which the frequency of dry circulation types was considerably larger than the frequency of wet types. The only exception was 2007, when the number of dry and wet types was comparable. According to two-sided Student's t test, the mean frequency of dry circulation types during VEG seasons when a drought occurred was significantly (at the 1% level) larger compared to the overall mean frequency of dry types in the data set in VEG.

For COLD, the positive trend of dry circulation types' frequency had smaller magnitude (less than 1% per decade) and was not significant at the 5% level (Figure 7b). The COLD seasons in the 2015–2018 dry period had average or even below-average frequency of dry circulation types compared to other seasons and therefore the 2015–2018 drought probably originated mainly from the dryness during vegetation periods (the fact that the link between droughts and dry circulation types during COLD was weaker compared to VEG, in general, could also play a role). Although the mean frequency of dry circulation types in COLD seasons with a drought was larger than the overall mean frequency of dry types in the data set in COLD, this difference was not statistically significant.

3.3 Dry Period of 2015–2018

In this section, the dry 2015–2018 period is analyzed in more detail by linking drought indices (averaged over the stations) to individual circulation types. Prevailing wet conditions of 2013, due to floods during June in central Europe (Blöschl et al., 2013), were succeeded by a transition to a drier regime by the beginning of 2014, as reflected in the long-term indices (PDSI and SPEI-52). Only a short-term drought was observed in VEG 2013 (Figure 8), which occurred in July and August simultaneously with heat waves (Lhotka & Kyselý, 2015). The shift to drier conditions was related to unusually large number of anticyclonic and southwesterly types (A, ASW, AW, and S), at the expense of cyclonic and northerly types (Table 3), resulting in low PZIN and SPEI-4 values in the beginning of 2014 (Figure 8).

This episode was followed by moister conditions at the end of the year, partially replenishing water availability before the extreme episode of 2015. The frequency of cyclonic (anticyclonic) types was higher (lower) in VEG 2014, compared to the previous COLD 2013/2014 season (Table 3). A noteworthy lack of cyclonic types occurred in COLD 2014/15 (no C, CNE, and CNW), preceding the major drought that developed in VEG 2015. All four drought indices showed extreme drought conditions in this season and this dry period was associated with exceptionally high number of dry circulation types (especially AN, ANE, ASW, ANW, and S), while the number of cyclonic types remained low (Table 3).

The extreme drought conditions persisted until mid-2016 (according to SPEI-52), or the end of 2016 (considering PDSI). The extreme drought was driven by unusually large abundance of dry ASW and AW types in COLD 2015/2016 and was related to the record-breaking high temperatures during this season in Europe (C3S, 2016). The largest discrepancy between the long-term drought indices was found during 2016–2017, when severe drought persisted according to PDSI, but only a short episode of moderate drought in mid-2017 was found using SPEI-52 (discussed in section 4). In 2016–2017, increased frequency of strongly anticyclonic (A) type was found in both VEG and COLD (Table 3), suggesting that advective processes were less important (compared to previous seasons) for maintaining dry conditions in this period (especially as captured by PDSI; Figure 8).

In the first half of 2018, both long-term indices dropped to severe drought level and drought conditions peaked in the second half of the year, associated with a high frequency of dry circulation types, comparable to that observed in the first half of 2015. The PDSI index reached a record-breaking value in 2018, but drought conditions were more extreme in 2015 according to SPEI-52 (Figure 8). Although the magnitude of the 2018 drought was comparable to 2015, its main driving mechanism was different. No increment in anticyclonic southerly to northwesterly types was observed (as in 2015) and the westerly advection was reduced in general (Table 3). By contrast, the frequency of dry northeasterly types (AN, ANE, AE, and E) was substantially heightened (together 52 days during VEG 2018, compared to the long-term mean of 30 days), indicating more frequent advection of dry continental air compared to previous years.