Soil drought and circulation types in a longitudinal transect over central Europe

2.1 Regions studied

Four regions of central Europe (Figure 1) were selected for the analysis of soil drought:

Climatological differences between the four regions analysed during the 1961–2019 period, based on temperature and precipitation data of the Czech and Slovak Hydrometeorological Institutes, appear in Figure 2. All four regions exhibit a simple annual temperature wave, with a maximum in July and with a minimum in January (Figure 2a). The mean annual amplitude increases steadily from west to east, from 20.1°C (CB) to 22.3°C (ES). The highest mean annual temperature is 10.2°C in SS and the lowest 8.9°C in both CB and ES (SM 9.6°C). Annual variation in precipitation totals is of continental distribution, with a maximum in June (SM, SS) or July (CB, ES) and with a minimum in February (CB, SM) or March (SS, ES) (Figure 2b). The highest mean annual precipitation total occurs in ES (681 mm), followed by CB (591 mm), SS (575 mm) and SM (545 mm). Based on the index of temperature continentality (Gorczyński, 1920), the four selected regions exhibit a growing continental nature from west to east (CB: 24.0%, SM: 27.3%, SS: 29.1%, ES: 30.0%). The index of precipitation seasonality, after Markham (1970), suggests that changes are already irregular (CB: 20.5%, SM: 23.1%, SS: 18.2%, ES: 22.2%).

2.2 Soil drought

Soil drought within the territory of the four regions defined above is described in terms of soil moisture data calculated from the SoilClim model (Hlavinka et al., 2011; Trnka et al., 2020). Working at a daily resolution within a network of 0.5-km × 0.5-km grids, this model calculates soil moisture for depths of 0–100 cm. The calculation requires maximum and minimum air temperatures, global solar radiation, precipitation total, vapour pressure and wind speed as meteorological inputs, as well as data concerning soil properties and vegetation cover. The soil/water balance modelling includes the snow-cover aspect. Comparison of drought anomalies between areas with different levels of mean soil moisture was enabled by converting data calculated from SoilClim into percentiles for the 1961–2010 period, created individually for every grid point, date and soil profile.

The model has been extensively tested by Trnka et al. (2015a; 2015b; 2020) and has been able to explain between 74 and 80% of the daily actual evapotranspiration variability measured by eddy covariance and Bowen ratio systems over three sites, six crops and multiple seasons. Also differences between three soil types have been well captured by the SoilClim using the lysimetric station Hirschstetten (Austria) explaining up to 63% (topsoil) and 74% (subsoil) of observed daily soil moisture variation with RMSE ranging from 2.8 to 4.2% for both layers in the 1999–2004 period. SoilClim has been shown to reproduce well changes in the long-term soil moisture dynamics in the topsoil, especially during the April–September periods, as well as replicate trends reported from pan evaporation measurements taken between 1968 and 2010 from five representative stations across the Czech Republic (Trnka et al., 2015b).

In order to address actual soil drought, values at the ≤10th percentile were further employed. The gridded data were then aggregated for each of the four regions, using the percentage of grid points fulfilling the 10th-percentile drought condition for every day. The value of the 10th percentile has been used as severe drought indicator, for example, by US Drought Monitoring (http://droughtmonitor.unl.edu/) since 1995 (Svoboda et al., 2002) and adopted also by the Czech Drought monitoring system (www.interdrought.cz).

2.3 Precipitation

Daily precipitation totals from ~1,100 rain-gauge stations of the CHMI and Slovak Hydrometeorological Institute (SHMI) were also used as stand-alone indicators and were calculated for 0.5-km × 0.5-km grids for the entire 1961–2019 period. These data were again aggregated for the four selected regions using the arithmetic mean of grid points included.

2.4 Circulation types

Atmospheric circulation over the four regions was analysed in terms of circulation types derived from three circulation indices: flow strength, flow direction, and vorticity (Jenkinson and Collison, 1977; Plavcová and Kyselý, 2011). These were calculated at daily resolution using sea-level pressure (SLP) at 16 points centred over individual regions, at the following coordinates for the centres: CB: 50.36°N, 15.16°E; SM: 48.89°N, 16.90°E; SS: 48.06°N, 17.96°E; ES: 48.82°N, 21.81°E (see the location of 16 SLP points using CB and ES regions as examples that appear in Figure 3). The SLP data for calculation of circulation indices for the 1961–2019 period were taken from the NCEP/NCAR reanalysis (Kalnay et al., 1996). Because the spatial resolution of the NCEP/NCAR reanalysis' grid (2.5° × 2.5°) did not match the coordinates of the SLP points, a weighted average of the nine nearest reanalysis' grid points (weighted by their inverse distance from the respective SLP point) was used.

Based on the weather types suggested by Lamb (1972), a total of 27 circulation types were objectively classified through the scheme developed by Jenkinson and Collison (1977), using the STR, DIR and VORT indices. If both STR and VORT are lower than 3, a given day remains unclassified (type U). When VORT is at least four times larger than STR, a pressure field is classified as of strongly cyclonic (type C, if VORT >0) or strongly anticyclonic (type A, if VORT <0). If STR is larger than the absolute value of VORT, those days are classified into one of eight directional types based on DIR (N, NE, E, …, NW). Finally, the remaining days are assigned to one of hybrid types based on their DIR and VORT (AN, ANE, …, ANW; CN, CNE, …, CNW). The classification algorithm is summarized in Table 1 and individual circulation types were visualized in Figure S1 in Supplementary Information.

Because the method was originally developed to represent circulation patterns over the British Isles, an area characterized by larger pressure gradients compared to central Europe, slight modifications had to be made to its parameters in order to avoid a scarcity of directional types and a surplus of unclassified days (Plavcová and Kyselý, 2011). Although this adjustment was partly subjective, the resulting circulation types were clearly linked to daily maximum and minimum temperatures (Plavcová and Kyselý, 2012), precipitation (Plavcová et al., 2014), and also heat waves (Lhotka et al., 2018) and droughts (Lhotka et al., 2020).

4.1 Spatiotemporal variability of soil-drought episodes

The spatiotemporal variability of soil drought in the areas studied appears in Figure 5, expressed as mean percentages of grid points with 10th-percentile drought in the summer and winter half-years during the 1961–2019 period. The figure provides generally clear agreement in the occurrence of individual episodes, with varying intensities of soil drought reflected in the four regions. Patterns of opposites appear over time in the two half-years: while soil-drought episodes in the summer half-year are predominant in the second half of the period studied (particularly 1990, 2003, 2007, 2012, 2015 and onwards), such episodes in the winter half-year occur mainly in 1961–1990 (e.g., 1962/1963, 1969/1970, 1983/1984, but also 2011/2012). This contrast also appears in values for linear trends, which are positive in the summer half-year for all four regions (for CB and SM statistically significant) and negative in the winter half-year (Table 3). Individual dry episodes in the four regions coincide with one another to different degrees. As expected, the highest Pearson correlation coefficients are to be found between the closest regions – SM and SS: 0.72 in the summer half-year and 0.69 in the winter half-year. The weakest correlation in the summer half-year (0.32) is, surprisingly enough, between the two Slovak regions (SS and ES). In the winter half-year, no correlation relationship exists between CB and SS (−0.01), while all other correlation coefficients are at least statistically significant.

4.2 Precipitation related to drought episodes

For a given region, precipitation totals for the three individual phases of soil-drought episodes were expressed as percentages of their mean totals for the 1961–2019 period. The values for all four regions appear in Figure 6. The lowest percentages occur in Phase I, leading to corresponding soil-drought episodes (from 18.3% in SS to 26.1% in ES). This is partly an artefact the delimitation of Phase I, beginning when precipitation deficit starts to develop, but it still demonstrates the existence of short, extremely dry periods before the onsets of soil-drought episodes. These percentages almost doubled during the course of drought episodes in Phase II (from 35.7% in CB to 43.1% in ES). In Phase III, the sudden termination of the previously-defined episodes is clearly demonstrated in high precipitation increases, achieving between 227.7% in ES and 256.3% in SS; both Czech regions exhibited increases of approx. 250% in mean precipitation.

4.3 Drought episodes and circulation types

4.3.1 Frequency of circulation types

The mean relative frequencies of individual circulation types for the 1961–2019 period show very similar patterns for all four regions (Figure 7), although differences between them may reach up to 3% (type W) in some cases. The most frequent circulation type in all regions is A, followed by AW (SM, SS, ES) and W (CB); the third most frequent types were AW (CB), W (SM), SW (SS) and S (ES). Grouping the types, the most frequent were anticyclonic (from 42.7% in CB to 46.6% in SS), followed by directional types (from 35.6% in SS to 38.9% in CB), while the least frequent were cyclonic types (from 15.3% in ES to 17.2% in CB). Between 1.3% (CB) and 2.0% (SS) of days remained unclassified.

Table 4 shows the circulation types with the relatively highest positive and negative deviations from their 1961–2019 means in Phases I–III of soil-drought episodes in the four regions analysed, from which follows:

1. In the origin of soil-drought episodes (Phase I), increases in frequency of types ANE (all regions) and A (except ES), as well as AN in both Slovak regions, were notable. The greatest decreases in frequency were recorded for types W (all regions) and SW (except CB). This also appeared for type AW in both Czech regions and for type C in both Slovak regions.
2. In Phase II of soil-drought episodes, increases in the frequency of type AN (except ES) were notable; the same held for type ANE for both Slovak regions. On the other hand, the highest negative deviations appeared for type SW (except SS). Also worthy of note is the relation between the closest regions, SM and SS, with positive deviations for type A and negative for type C. Surprisingly, the regions most distant from one another, CB and ES, exhibited important positive deviations for type AE and negative for type W.
3. In the end Phase (III) of soil-drought episodes, a total of eight circulation types with the highest positive deviations occurred in four regions, of which type SW was present in all of them except SM. Two other types were recorded in two regions: W in both Czech regions and C in SM and SS, which lie very close to one another. Turning to the highest negative deviations, type A proved important in all regions except SS; while type AE was influential in SM and SS, as was type ANE in SS and ES.

TABLE 4. Circulation types with the highest relative positive (P) and negative (N) deviations from their 1961–2019 means in Phases I–III of soil-drought episodes (Phase I – origin, Phase II – course, Phase III – end) for the CB, SM, SS and ES regions

Region	Phase I				Phase II				Phase III
	P		N		P		N		P		N
	Type	%	Type	%	Type	%	Type	%	Type	%	Type	%
CB	SE	4.7	W	−4.0	AE	2.7	W	−4.8	W	5.5	A	−4.7
	ANE	4.1	NW	−2.9	AN	2.5	SW	−2.3	NW	5.0	ANW	−3.2
	A	2.7	AW	−1.5	E	2.2	AW	−2.3	SW	3.4	SE	−3.1
SM	A	5.8	W	−3.0	A	2.8	SW	−1.9	W	5.4	A	−5.5
	AE	4.4	SW	−2.9	ANW	2.6	C	−1.5	C	3.5	ASW	−2.7
	ANE	3.0	AW	−2.7	AN	2.2	NE	−1.4	E	2.4	AE	−1.9
SS	A	7.2	SW	−2.8	A	4.1	C	−1.5	SW	4.7	AE	−3.4
	ANE	4.3	C	−2.1	ANE	1.8	CSE	−1.1	U	4.7	ANE	−3.2
	AN	4.0	W	−2.0	AN	1.6	E	−1.0	C	4.3	N	−3.1
ES	AN	3.8	SW	−3.6	SE	2.4	W	−2.5	SW	6.1	ANE	−3.4
	ANE	3.6	W	−2.3	AE	1.4	SW	−1.6	CW	3.0	A	−3.3
	NE	2.6	C	−1.8	ANE	1.4	ASW	−1.2	AN	3.0	ASE	−3.1

Summarizing these results, increased frequencies of anticyclonic types ANE, A and AN, and decreased frequencies of directional types W and SW, as well as of the cyclonic type C, appear the most influential in the origin and course of soil drought in the whole study area.

Partly similar patterns over the four regions appeared in groups of circulation type during the three phases of soil-drought episodes (Figure 8). Patterns leading to their origin (Phase I) are typified by increased frequencies of anticyclonic types, while frequencies of cyclonic and directional types exhibit negative deviations. The dominant deviations in all three circulation groups appear in SS followed by SM, while for CB are the smallest. Lower deviations of all three circulation groups with the same signs also occurred during the course of soil-drought episodes (Phase II). Differences between individual regions were considerably less marked than in the previous Phase I and deviations were higher for the two Czech regions than for the Slovak ones. While the highest deviations in the three circulation groups occurred for SM and CB, the lowest always appeared for ES. The end of soil-drought episodes (Phase III) was related to marked decreases in the frequencies of anticyclonic types and to increases in cyclonic types. Directional types exhibited an increase for CB and SM, while the deviations in frequencies in the two Slovak regions were slightly negative. Deviations for CB were the highest in each circulation group (only for cyclonic types comparable with SS).

5.1 Spatiotemporal context and data uncertainty

The definition of drought episodes relied on use of the SoilClim model (Section 2.2). This simplified water-balance model uses daily weather data collected by the CHMI and SHMI in a station network of ~1,100 rain-gauge stations and more than 400 climatological stations. The mean distance between neighbouring stations is approximately 22–30 km for climatological stations and approx. 10 km for rain-gauge stations. A locally-weighted regional regression, taking elevation and other terrain characteristics into account, was employed to interpolate the daily data into 0.5-km × 0.5-km grids. Although the daily global radiation balance considers slope, aspect and horizon obstruction (Schaumberger, 2005), uncertainty related to the nature of the ground data cannot be completely ignored. The soil-moisture content, as estimated by the SoilClim model (Hlavinka et al., 2011; Trnka et al., 2020), is principally based on the modelling approach suggested by Allen et al. (1998). Proper representation of not only soil-water holding capacity within each grid but also of the type of vegetation cover, phenological development, root growth and snow-cover accumulation, sublimation and melting (Trnka et al., 2010) also have their effects on the onset and duration of individual drought events. The module for actual evapotranspiration and soil-water content simplifies otherwise complex soil water transport processes by using only two soil layers (0–40 cm and 40–100 cm). Such a cascading approach may lead to a lag of 1 or 2 days in estimating onset and end of any particular event. SoilClim estimates of the soil-moisture content are affected by the maximum soil-water holding capacity for the two soil layers in each grid cell. Within grids in which at least some part of the growing season is influenced by high underground water tables (which are likely to be reached by roots during natural subsurface irrigation), and that therefore respond to drought differently (both in terms of stress magnitude and timing), the higher availability of soil water is accounted for. Soils with an observed gleyic process in close proximity to major water bodies, peat, and bog areas had significantly slower soil-moisture depletion rates than neighbouring grids lacking such influences (Trnka et al., 2015a). The soil moisture anomaly (i.e., a percentile-based value) has been used herein when defining the onset and the end of any given drought event. This approach limits the effect of uncertainties in the soil and terrain conditions. Further, the selection of the regions for study took into consideration the higher uncertainty of drought estimates in complex terrain and focused on comparatively flat and homogenous regions (see Figure 1).

5.2 Soil-drought episodes and circulation types

The objective classification used herein proved capable of presenting a number of clear patterns, including differences between the regions, in terms of individual circulation types with respect to precipitation and soil drought occurrence. The CB region appeared, in general, the most markedly different from the others, while westerly airflow over central Europe was significantly more involved in precipitation than it is in more eastern regions. Further, directional circulation types exhibited the most significant changes in frequency during all phases of soil-drought episodes in CB, frequently with contrasting effects (e.g., a strong increase in frequency of the SE type and a decrease in the W type during Phase I). In addition, more precipitation fell during directional types in CB than in the other regions, followed by ES, where it probably appears to be a matter of windward/leeward effects rather than of any west–east gradient. However, a west–east gradient is reflected in terms of precipitation in certain airflow directions: types with southerly airflow are more important in the eastern regions, at the expense of the western type.

A possible relationship between atmospheric circulation and drought occurrence in the Czech Republic has already been investigated by Trnka et al. (2009), who used the Hess–Brezowsky catalogue of Grosswetterlagen (Hess and Brezowsky, 1952) as circulation indices, together with a wide range of meteorological drought indices for the summer half-year, and found anticyclonic, easterly and southerly types to be associated with drought occurrence. This tallies in part with the current article, except in the matter of southerly types: this article identifies hybrid anticyclonic or north-eastern circulation types as more important drought drivers. However, it must be borne in mind that the circulation types used by Trnka et al. (2009) are more representative for Bohemia than for regions located farther east. Even so, the increasing trend in the frequency of drier circulation types as defined by the Hess–Brezowsky catalogue since the end of the 19th century pointed out by Trnka et al. (2009) corresponds with a significant positive trend in the anticyclonic types of objective circulation classification used herein (Table 6).

The occurrence of anticyclonic types and advection from the eastern quadrant in general is associated with an interruption of zonal flow into Europe. Many authors have suggested that rapid warming in the Arctic in recent decades (Arctic Amplification, see Screen and Simmonds, 2010) and related lower meridional temperature gradients may affect the strength of westerlies over the Northern Hemisphere (e.g., Francis and Vavrus, 2012; Tang et al., 2013). It should be noted, however, that more recent studies have pointed out that complex processes over the Arctic present substantial challenges to robust signal detection (Francis et al., 2017) and thus the links between Arctic Amplification and mid-latitude atmospheric circulation remain uncertain (Coumou et al., 2018). The observed increase of anticyclonic types in central Europe (Lhotka et al., 2020) cannot, therefore, be reliably attributed to changes in the Arctic.

Řehoř et al. (2020) employed another subjective classification of synoptic types, developed by synoptic meteorologists of the CHMI (Kolektiv pracovníků synoptické a letecké služby HMÚ, 1967), to analyse 10th-percentile soil-drought episodes in four small areas of the Czech Republic in the 1961–2017 period, of which only one region partly overlaps with SM in the current study. Despite the fact that soil-drought episodes were not analysed separately for Phases I–III, Řehoř et al. (2020) did not find the same association with northerly/easterly airflow, but a range of effects of westerly airflow on different regions as important soil drought drivers, which partially agrees with the current article. An advantage of the CHMI classification was its capacity to define very specific types as troughs of low-pressure over central Europe, but suspected non-homogeneity in the classification, absence of distinction between purely cyclonic situations and situations with only a frontal zone encroaching upon central Europe (herein, directional types) and the impossibility of modifying the classification for different regions must be noted among its drawbacks. Because Phase I (as herein) is most significantly associated with northerly/easterly airflow, absence of analysis of circulation before the onset of soil drought in Řehoř et al. (2020) is a possible reason for the absence of drought connection to the northerly/easterly airflow in that study, together with its different classification.

The same type of circulation classification as that employed herein was also used by Lhotka et al. (2020) for the Czech Republic, with the classification centred on Bohemia, although meteorological drought indicators were used rather than soil-moisture percentiles and the period investigated was 1948–2018. As emerged in Trnka et al. (2009), easterly airflow was disclosed as a drought driver, although accompanied by a southerly direction, not northerly as herein. This indicates that this perceived difference in crucial airflow direction may also be influenced by the use of either meteorological drought indicators or soil-moisture percentiles for the drought categorisation.

5.3 Trends in soil-drought episodes

Previous investigations of soil-moisture trends have provided a mixed picture. Trnka et al. (2015b) revealed more significant trends in soil-moisture decrease in the summer half-year for the 1961–2012 period in the area of the CB region compared with SM, which agrees with the increasing trends in soil-drought frequency herein. On the other hand, Řehoř et al. (2020) found the steepest trend of soil-drought frequency increase in 1961–2017 period in north-eastern Moravia and the least, gradual, one for north-eastern Bohemia, a region neighbouring the CB region delineated herein. This may suggest both that differences in soil-drought frequency trends may often be focussed as finely as genuine local differences (e.g., orographic), rather than as larger-scale gradients and that trend value is significantly influenced by extreme values at the beginning and at the end of the time series used, as was the case for the extreme drought of 2018–2019 in CB.

Trnka et al. (2015a) identified temperatures and precipitation as perhaps the most important factors influencing, among other things, the values of soil-moisture/drought. While precipitation is only source of water for the soil, temperatures contribute directly to its reduction through evaporation; prolongation of the vegetation season also contributes to higher transpiration, while a shorter freezing period leads to a reduction in the snow accumulation period. In terms of linear trends for the 1961–2019 period (Table 6), annual, summer and winter half-year temperatures exhibit statistically significant increases: for annual series of between 0.34°C·10 years⁻¹ in SM and 0.39°C·10 years⁻¹ in SS. In the summer half-year, the trends lie between 0.37°C·10 years⁻¹ in SM and 0.43°C·10 years⁻¹ in SS, while in the winter half-year they are between 0.31°C·10 years⁻¹ in SM and 0.38°C·10 years⁻¹ in ES. These results tally closely with the results of Zahradníček et al. (2020), who analysed linear temperature trends in the Czech Republic for the same 1961–2019 period. Although their trends are slightly higher in the summer half-year, this difference cannot explain the significant difference in soil-drought trends between the two half-years indicated in Table 3. However, the secondary effects of overall increased temperatures, leading both to more water infiltration and evaporation in the winter and consequently greater water deficit in the summer half-year, appear to be the main drivers of this difference in soil-drought trends. The west–east gradient of soil-drought occurrence trends does not appear to arise out of temperature trends.

Precipitation totals exhibit insignificant linear trends in annual series that progressively increase from west to east, from −5 mm·10 years⁻¹ in CB to 17 mm·10 years⁻¹ in ES, with a similar situation in both summer and winter half-years (Table 6). Even though the trends are insignificant, they correlate with the west–east gradient of soil-drought occurrence trends (see Table 3). It is possible that the driver of the increase in soil-drought frequency in the summer half-year intensifies progressively in a west–east direction and it is statistically significant only in the Czech regions, but not in the Slovak ones. The statistical insignificance of the trends however, may suggest that this is merely an internal variability effect, not a long-term change.

Linear trends in the frequencies of groups of circulation types during the 1961–2019 period show consistent behaviour in the four regions (Table 6). Statistically significant increasing trends in the anticyclonic types and decreasing trends in the cyclonic types appear in the annual and summer half-year series in all regions (significant increasing trends in the winter half-year were evident only for anticyclonic types in SS). Insignificant negative trends also occur uniformly for the directional types. Trends in circulation types are in agreement with increasing soil dryness during the summer half-year and insignificant changes in soil drought occurring during the winter half-year, as shown in Table 3.