Vulnerability of hop-yields due to compound drought and heat events over European key-hop regions

1.1 The compound risks of extreme CEs and hop-yields losses

The dual concept of crop losses and compound CEs has been suggested as an approach to understand extreme impacts and reduce farmers' exposure to weather-related financial risks. However, a weakness of the CEs approach is the challenge of developing parametric models that reliably capture the yield losses experienced by farmers (MoU, 2018). With significant climate warming, droughts over European hop-growing regions are drawing increasing interest. Hop production can be an example for demonstrating the wide-ranging effects of compound events when the impacts depend on multiple dependent climate variables. Hop (Humulus lupulus L.) is a globally important crop and is grown especially for its secondary metabolites (e.g., hop oils, α- and β-acids), which are used in beer brewing to add bitterness and aroma to beer (Colen and Swinnen, 2016). Hop yards are exposed to an ensemble of CEs whose impacts are complex and difficult to assess (Potop, 2016; Potopová et al., 2019b). Figure 1 summarizes the impact of compound wet-cool and dry-hot thresholds on hop production, although each of these individual conditions may not necessarily be extreme in and of itself. Both extreme dry-hot and cool-wet thresholds lead to lower alpha–acid contents and can also decrease the yield of hop cones. Exposure to strong wind, for example, can cause leaf damage and some loss of cone-bearing laterals, both of which affect bine health and yield. From flowering to maturity, exposure to hot winds can also negatively affect cone quality. Simultaneous heat waves and drought events (i.e., a period of consecutive extremely hot days and low precipitation) lead to the expansion of weeds, many of which are C4 plants (e.g., panicoid weedy grasses—Echinochloa, Setaria, and Digitaria and some broadleaved species such as Amaranthus retroflexus and Amaranthus powellii), and have a higher water use efficiency compared to C3 plants (Rybáček et al., 1980; Jursík et al., 2018; Procházka et al., 2018). These CEs in turn contribute to increases in insect populations; hop cultivars that have been weakened by drought are generally less resistant to insects and other pests. Furthermore, storms, hail and longer rain periods with high air humidity lead to damaged plant tissues, stressed plants, and easier infection by fungal diseases (e.g., Pseudoperonospora humuli). The occurrence and distribution of hop peronospora is very closely related to concurrent variations in air temperatures, relative air humidities and precipitation amounts. These pests and diseases cost overseas producers millions of euros annually in crop loss and crop protection expenses. In previous years, the susceptibility of the hop crop to a range of pests and diseases has significantly affected market prices (IHGC, 2018). Rising demand and lower yields have driven the increases in hop prices over the past decade (EU report, 2018).

Climate change affects three of beer's core ingredients, namely, hops, water, and barley (e.g., Almaguer et al., 2014; Nelson, 2014). Therefore, the impacts of CEs on hop production and the adverse impacts of climate change has been recognized as one of the greatest challenges of the 21st century (e.g., Xie et al., 2018). Xie et al. (2018) assessed the vulnerability of the global beer supply to disruptions by extreme drought and heat events that may occur during the 21st century. Consequently, following this reasoning, the main motivation of this study was to call attention to the risks from increasing frequency of concurrent drought and heat events for the hop production. There have been no previous complex studies of the interactions between climate CEs and hop yield quantities in the main European hop-producing countries. The majority of studies have been developed by addressing individual factors, even though the impacts usually depend on multiple potentially dependent variables (Cerenak et al., 2010; Mikyška and Jurková, 2011; Gloser et al., 2013; Krofta et al., 2013; Srečec et al., 2013; Čeh, 2014; Kolenka et al., 2016). With regard to our study area, the majority of hop studies have been focused on new trends in hop brewing, beer and the hop market in Germany (e.g., Weihrauchet et al., 2012), Czech Republic (e.g., Hejnák et al., 2015; Natsume et al., 2015; Patzak et al., 2015; Donner et al., 2020), Great Britain (e.g., Darby, 2006; Darby, 2010; Capper and Darby, 2014), France (e.g., Steyer et al., 2014), Spain (e.g., (Cameira et al., 2007; Fandiño et al., 2015) and Slovenia (e.g., Pavlovič, 2012; Pavlovič et al., 2012; Pavlovič et al., 2013; Pavlovič, 2014). In all hop-growing regions across Europe, extreme CEs limit beer supply, and hop prices are strongly influenced by the weather and by the level of stocks (Colen and Swinnen, 2016).

1.2 The key hop-growing regions in Europe

The hop plant is grown throughout most moderate climate regions of the world. In the Czech Republic, the growing season of hops (from bud burst to cone development, GS) takes 102 days on average with a sum of air temperatures of 1,537°C, a sunshine duration of 731 hr, and a total rainfall amount of 176 mm with 28.5 total days of precipitation of at least 1 mm (Hájková et al., 2012). Europe is the second largest beer producer in the world (Kirin Beer University Report, 2019). Some 2,600 farms in the EU grow hops, covering a total of 26,500 ha, which represents 60% of the total worldwide surface area used for hop growing. Fourteen EU member states produce hops, but the top producers include Germany (DE) and the Czech Republic (CZ), which together account for over 80% of total EU production. Our study includes the key hop-growing regions across the EU (Figure 2) such as Germany (Hallertau, south-eastern DE); the Czech Republic (Úštěcko, Žatecko, Tršická, Polabí lowland and suburban regions of Praha); Great Britain (Kent, south-eastern UK); France (FR, Alsace); Spain (León and Galicia, north-western ES); Slovenia (SI, Koroška); and Poland (Lublin, eastern PL). The largest hop-growing areas include the Hallertau region in DE and the Žatec region in the CZ. Alsace is by far the main hop-producing region in France. The largest harvested area is in DE, followed by CZ, UK, PL, SI, ES and FR (Figure 3). The total hop acreage is decreasing steadily in the EU, with a 12% reduction since 2001. Between the late 1990s and early 2000s, excess hop production caused a price depression. Nevertheless, with the high price volatility in the commodity markets after 2007, the hop prices increased again with hops again becoming a lucrative cash crop. Bitter varieties are grown in approximately one-third of the area. The largest holdings are in CZ (40.7 ha/holding) and the smallest are in Spain and Poland (around 2.0 ha/holding; Pavlovič, 2012).

1.3 Objectives of the study

To identify the multivariate adverse weather events that are associated with extreme impacts on the commercial production of hops is challenging and is often made even more difficult due to small sample sizes in observations of the contents of alpha-bitter acids over the European Union (EU). The purposes of the present study are therefore (a) to provide a frequency analysis of the univariate and bivariate extremes of drought and heat waves in the main European hop-growing regions; (b) to determine the risk of occurrence of the four extreme compound modes, namely, dry-cool, dry-hot, wet-cool and wet-hot, for each month of the hop-growing season using the copula technique in various climatic conditions and (c) to estimate the risk of lower yields from the hop cones based on concurrent dry-cool, dry-hot, wet-cool and wet-hot modes over the target areas.

2.1 Climate-hop datasets

Country-level statistical data of harvested areas and hop yields were downloaded from the FAOSTAT database for all of the seven studied hop-growing countries for the 1961–2018 farming years. To more deeply analyse hop yields, we used the traditional Czech hop-growing regions as a specific case study: Žatec, Tršice and Úštěk with 73.8, 16.6 and 12.5% of the total hop yard area in CZ, respectively. The statistical analysis of Czech hop production was conducted using the harvested areas (ha), and yields (tha⁻¹) of Saaz semi-early red bine hop (tha⁻¹) during the period 1961–2018. Saaz hops are fine semi-early aromatic hops used by breweries around the world because of their unique characteristics. In the brewing industry, especially in the production of high-quality beer brands, the Saaz hop plays a very important role. By using Saaz hops, a beer with a delicate and soft hop aroma and a balanced pleasant taste can be produced. The composition of hop resins is specified by the relatively low content of α-bitter acids, which is in the range of 2.5–5.5%.

We selected the European Climate Assessment Dataset (ECA&D), which is mostly based on station-observed data (Klein Tank et al., 2002), for the following stations: Eichstatt-Landershofen (DE), Strasbourg-Entzheim (FR), Goudhurst (UK), Wlodawa (PL), Santiago De Compostela/Labacolla (ES), Leon Virgen Del Camino, Spain (ES) and Celje-Medlog (SL). Both daily maximum temperatures (T_max) and precipitation (P) datasets are used for detecting droughts and heat waves and our CEs modes. For CZ, we further used daily T_max and P from the observational dataset of the Czech Hydrometeorological Institute (CHMI) for the Tuhan, Doksany, Kněževes, Olomouc, Žatec, Poděbrady and Praha-Ruzyně stations (Table 1).

2.2 Detection of univariate and bivariate extremes of droughts and heat waves

To provide reliable risk measures for hop production, we argue that defining extreme climatic events with respect to their impacts is important. The different responses to concurrent extreme drought and heat events in contrast to univariate heat-only events and their probability of occurrence during the GS have been evaluated. Events in which only one variable is extreme (e.g., either hot but no drought or extreme drought but not hot) and events in which both variables are extreme (e.g., drought and heat waves) may have different impacts on hop yields and alpha-bitter acid contents. Simultaneous CEs and single-factor extreme events were detected using the following stepwise procedures: (a) only-hot (days suffering from heat stress, HD); (b) only heat waves (HW); (c) accumulated heat stress during HW (TS30 index); (d) concurrence of two extreme climate variables dry-hot (DHD); (v) drought-heat waves (DHW); and (e) TS30 index during DHW. We used compound extreme joint modes of the 85 and 35% percentile thresholds to define the adverse effects of precipitation and temperature on hop yields. A T_max85/P35 mode denotes temperature events above the 85% percentile threshold and precipitation events below the 35% percentile threshold (Zhou and Liu, 2018).

To define a standardized and comparable method for HWs across all sites, we selected a 3-day heat wave with an 85th percentile threshold, which was calculated as three consecutive days with T_max exceeding the 85th percentile of the long-term climatology within the GS for that station. An example of the definition of HWs during GS, is shown in Figure S1a. To identify the heat stress intensity during HWs and DHWs, TS30 indexes were calculated as the accumulated heat degree days when the daily T_max≥30°C. We argue that this method is well suited to investigate heat waves for the various hop-growing regions over Europe. A higher threshold for T_max may result in very few events being recorded in the UK, PL, DE and CZ hop-growing regions and a smaller threshold may result in too many events being recorded in FR, SI and ES. The univariate HW and bivariate DHW approaches are distinguishable by their core formation mechanisms (Huth et al., 2000; Mueller and Seneviratne, 2012; Zhang et al., 2019). Co-occurring of drought and heat stress reveals that the stress combination has several unique aspects, combining high respiration with low photosynthesis, closed stomata and high leaf temperature (Mittler, 2006). Meanwhile, as soil and vegetation dry out, land evapotranspiration is reduced, hence the air becomes even drier, which may further decrease the likelihood of rainfall and favour the occurrence of droughts (Miralles et al., 2019). Concurrently, as evaporation progressively declines, a larger fraction of incoming radiation is employed to warm up the environment, which leads to an accumulation of sensible heat in the atmosphere that may develop into a HW. For example, according to the results by Miralles et al. (2014) 40% of the heat stored in the atmospheric boundary layer during the 2010 M-heatwave came from advection.

2.3 Hop-yield anomalies as a function of bivariate return periods

We used the concurrent bivariate return period to model the joint probability distributions of precipitation-temperature extremes and to provide risk assessments for concurrent droughts and heat waves during the hop-growing seasons. The concept of copulas is widely used in hydrological research, but little attention has been paid to the risk analysis of concurrent droughts and heat waves (Schölzel and Friederichs, 2008; De Michele et al., 2013; AghaKouchak et al., 2014; Zhou and Liu, 2018). Zscheischler et al. (Zscheischler et al., 2017) explored how closely the bivariate copula of temperature and precipitation are linked to crop yields. Zhou and Liu (Zhou and Liu, 2018) applied the concurrent bivariate return period methodology based on the concept of copulas designed to model the dependence between multiple variables. In a similar fashion, we use bivariate return periods for a given temperature–precipitation couple (x, y) to investigate how their anomalies are related to hop yields. First, if we assume the two variables X (precipitation, P) and Y (maximum temperature, T_max) with cumulative distribution functions F_x(x) = Pr(X ≤ x) and F_y(y) = Pr(Y ≤ y), then the copula (C) can be used to obtain their joint distribution function:

(1)

where F(x, y) is the joint distribution function of X and Y, that is, F_x(x) and F_y(y) represent the marginal non-exceedance probability distributions of T_max (u = F_x(x)) and P (v = F_y(y)), respectively.

Second, from the joint distribution function, the so-called joint survival distribution can be obtained using the survival copula (Salvadori et al., 2013):

(2)

where and are the marginal survival functions of X and Y (i.e., u′ and v′ are the exceedance probabilities of T_max (u ′  = 1 − u) and P (v ′  = 1 − v)), and is the survival copula.

Third, the joint probability of the P35/T_max mode is applied (Zhou and Liu, 2018):

(3)

We subsequently computed joint return periods to estimate the risks (occurrence probabilities) of the four compound extreme modes: P35/T35, P85/T35, P35/T85 and P85/T85, which represent the combinations of extreme dry-cool, wet-cool, dry-hot and wet-hot events during the GS, respectively. The return level with a T year return period shows an event that has a 1/T chance of occurrence in any given year. Joint return periods can be investigated depending on the selected copula. For any given (x, y)ɛ R², there exists a unique critical survival layer (or isoline) at which the set of realizations of X and Y share the same probability t ɛ(0, 1) (AghaKouchak et al., 2014):

. The critical region is calculated as (De Michele et al., 2013). For bivariate copulas, the return period (RP) is calculated as follows:

(4)

where μ is the average inter arrival time of events and is the critical region using survival functions and survival copulas.

Before modelling the joint probability distributions of dry-cool, wet-cool, dry-hot and wet-hot events, we selected four copulas. To capture events with low-return periods and not just the extremes, we selected the most prominent bivariate Archimedean copulas (Gumbel, Clayton and Frank asymmetric theoretical probability distributions) and the t copula (symmetric theoretical probability distributions; the upper tails are dependent). When modelling with copula operators (R Core Team, 2018), Kendall's tau rank, the Kolmogorov–Smirnov test (K–S; Equation 5), least root-mean-square error (RMSE; Equation 6) and Akaike information criterion (AIC; Equation 7) are first applied to determine the best-fit distribution.

(5)

where (x_i, y_i) (i = 1, 2, …, n) is the observation sample, F₀(x_i, y_i) is the joint distribution of sample estimates, m_i is the number of observed samples that satisfy the conditions of x ≤ xi, y ≤ yi, n is the total number of observed samples.

(6)

(7)

where Oe_i and O_i are empirical frequency and theoretical frequency of the i-th compound sample, respectively, m is the number of the joint distribution function parameters, n is the number of samples.

The selected distribution (t probability distributions) or copula family is rejected if the p-value is less than .05. This occurs at less than 5% of datasets for each hop region, which is in the acceptable range of tests that may fail by random chance (Zscheischler et al., 2017). The Frank copula is suitable for modelling dataset characterized by weak tail dependence (Table 2). In our study, the Frank copula permits the approximation of the upper and the lower tails, and thus the Frank copula permits modelling positive as negative dependence in climate-yield datasets. The Gumbel copula can model only upper-tail dependence, whereas the Clayton copula can model lower-tail dependence. Hence, only using extreme value copulas might not be appropriate (De Michele et al., 2013; Zscheischler et al., 2017). After the marginal distributions of each variable are determined, we computed the occurrence probabilities for the four compound extreme modes by using Equation (4).

TABLE 2. Function expressions of Archimedean copulas and their parameters used in this study

Copula	Expression of bivariate copula, C(u, v)	Generator j (t)	Parameter space, upper and lower tails	Kendall's tau rank, relationship between t and a
Frank	– log [1 +	–log ()	α ≥ 0	τ = 1- (1- dt)
Gumbel	exp (− [(−log u)α + (−log v)α]1/α)	(−log (t))α	α ≥ 1	τ = 1–1/α
Clayton	(u–α + v –α – 1)–1/α	(t–α – 1)	α ≥ 0	τ = α / (α + 2)

The annual series of the crop yields at the national level from 1961 to 2018 were used to calculate the standardized yield residuals series (SYRS; Potopová et al., 2015). We then fitted regression models to explain the crop yield anomalies as a function of bivariate return periods using the natural logarithm of the bivariate return periods as a predictor (i.e., SYRS_hop = a + b ln(PR)) and only kept the model with the highest fraction of explained variance. Furthermore, we fitted ordinary linear models to each combination of hop-growing region and time period using P and T_max anomalies as predictors.

3.1 Variability of univariate hot days and bivariate dry-hot days in the key hop-growing regions

To determine how profoundly single events and CEs affect hop yield, we compared the variability of only-hot events and concurrent drought and heat from the period of bud burst to cone development in the key EU hop-growing regions (Figures 4 and 5) and for all CZ hop regions (Figures S1b-2) for the period 1961–2018. The HD and DHD occurrences show a meridional pattern, which gradually increases from the British and Polish hop-growing regions to large parts of the French, German, and Czech lands toward the Slovenian and Spanish hop sites. This pattern is important for hop production because it drives canopy growth patterns and the timing of flowering induction. The most frequent occurrences of concurrent dry-hot events tend to better align with those countries that have produced the most beer per capita in recent years. At the level of individual regions in each country, the highest frequency of dry-hot events occurred in Alsace, then in Hallertau (southeastern DE sites), Tršice and Polabí lowland (central and southern CZ sites), Koroška (northern part of SI), and Galicia (northeastern ES). The frequency of hot and dry-hot days for the CZ hop sites was higher than that for the DE hop regions. The British and Polish hop sites, with fewer hot events, have been found to exhibit greater interannual variability in the intensities of hot and dry-hot days, which leads to increased risk during the production life of hop fields.

The top three ranked GSs with the highest frequency of hot/dry-hot days occurred in 1976 (15/12 days), 1995 (13/10 days) and 2003 (8/5 days) in the UK. For the Polish hop areas, the highest occurrences of hot and dry-hot days were both seen in 2015 (18/16 days), 1994 (25/15 days) and 2010 (17/15 days), while for the CZ hop regions, the highest occurrences were in 2018 (32/27 days), 2015 (28/25 days) and 1994 (28/25). The top three ranked GSs with maximum DH/DHD occurred for DE in 2015 (30/25 days), 2018 (26/24 days) and 1994 (22/15 days); for France, the maximum DH/DHD occurred in 2003 (38/30 days), 2015 (34/29 days) and 2017 (27/22 days); the maximum DH/DHD for Slovenia occurred in 2003 (28 days), 2012 (25 days) and 2015 (25 days); and the maximum DH/DHD for Spain occurred in 1991 (41/40 days), 1962 (38/37 days) and 2017 (34/34 days). In 2015, the highest number hot and dry-hot days occurred in Alsace (FR), Hallertau (southeastern DE), Tršice and Polabí lowlands (central and southern CZ sites), Koroška (northern SI) and Lublin (eastern PL). The extreme year 2018 ranked at the top in the northwestern Czech hop regions (e.g., Úštěcko and Žatecko) and in all DE hop areas (Lhotka et al., 2018a; Lhotka et al., 2018b).

For all hop sites, besides the British hop areas, upward shifts toward drying and heat stress with the most change points in 1999 were significant at the 95% confidence level. After 1999, a change-point is related to an abrupt change (i.e., abrupt increase) in the distributional properties of HD/DHD time series. After 1999, the most pronounced increased frequency of joint dry-hot days was observed for hop regions in CZ (3.3–8.9% decade⁻¹) and DE (2.2–7.8% decade⁻¹) followed by SI (2.7–4.4% decade⁻¹), FR (1.7–3.5% decade⁻¹), ES (1.3–3.1% decade⁻¹) and PL (0.8–2.0% decade⁻¹). We then estimated the excess ratio (e.g., “speed of changes” of HD/DHD) between two separate periods, namely, an historical period (1961–1998) and the present (1999–2018) period. The strongest changes were seen in SI, DE, and CZ in which the HD/DHD has increased more than 2.9 times across the hop regions. Interestingly, moderate excess ratios (approximately one-fold) of HD/DHD have been observed across the hop regions of France and Poland, while slow changes have been observed in Great Britain and Spain. As a result, in the top hop-producing nations (e.g., DE and CZ), the speed of change in the occurrences of joint dry and hot events were 1.2 times higher than those in France, Poland and Spain during 1999–2018 farming years.

Figures 6, 7, S3, and S4) document the temporal distributions of days with heat stress and joint drought-heat stress and their intensities during the GS, as well as the smoothed dates of the first and the last hot day/dry-hot days in a given year over the key EU hop-growing regions (all CZ hop areas). The highest frequencies of severe hot days and dry-hot days were in the 2000s and 2010s. The most frequent occurrences of heat stress occurred during the beginning of flowering (21.5%, the last 2 weeks of July and the first week of August) in the UK and PL; the beginning of flowering–cone development in FR, DE, CZ and SI (59.3%, July–August); and from inflorescence emergence to cone development (66.9%, June–August) in ES. Early drought-heat stress occurred in the FR and ES hop regions; during the period of bud burst to first leaves, this stress occurred 3.3% of the time; from the beginning of side shoot emergence, this stress occurred approximately 5.3% of the time; at the time of inflorescence emergence, this stress occurred 30.9% of the time; and at the time of the beginning of flowering and cone development, this stress occurred approximately 55.9% of the time

3.2 Only-heat waves and drought-heat waves and their properties in the key hop-growing regions

We have investigated both only-heat waves and drought durations that co-occur with extreme temperatures (drought-heat waves) in the key EU hop-growing regions, as well as changes in intense events over the 1961–2018 farming years. The evolution of heat waves and drought-heat waves and their properties, such as severity and duration, is a key question for adaptation due to their impacts, including hop quality and quantity. Hop cones are directly exposed to exceedingly high temperatures and radiation during heat waves. A lack of protection against stress commonly leads to physiological disorders related to alpha content. Concurrences of all combinations of drought-heat wave intensities and durations have increased substantially in the western and central European hop areas. Figures 8 and 9 demonstrate the interannual variability in the number of heat waves and the accumulated heat stress during the heat waves in the hop-growing regions of Hallertau (DE), Alsace (FR), Lublin (PL), Leon and Galicia (ES), and Koroška (SI). Figure 10 summarizes the heat wave frequencies and distributions during the GS for each traditional CZ hop region (e.g., Žatec, Úštěk and Tršice), as well as the Polabí lowland historical hop region and for a potential new production area. Drought-heat waves occur between May and September with a maximum frequency in July and August, although long and severe drought-heat waves in the UK and PL are relatively rare. The maximum number of heat wave days ranges between 15 days for the British and Poland hop areas, nearly 20 days in Central Europe, and more than 32 days in the southern European hop areas. For the British and Polish hop areas, the most heat waves lasted between three and 7 days, while durations longer than 10 days were rare. In 1976, the UK hop areas experienced the longest heat wave duration and joint drought-heat waves, and the severity was exceptional over most of the British Isles, France and Eastern Europe (Russo et al., 2015). An exceptional heat-wave lasted for mid-June to mid-July with T_max not falling below 32°C, while drought durations lasted for up to 16 months. Although the drought peak was in summer, 1976, low precipitation during the preceding winter exacerbated the extreme precipitation deficits that occurred in spring. Further major heat waves/drought-heat waves were found across the UK hop regions (2003, 1995, 1983 and 1990 sorted by severity) but all them were less pronounced than in 1976. In Alsace, the drought-heat wave of 1976 was also ranked at the top due to its extreme length, but this event had a lower severity than the event that occurred in June–August of 2003 (e.g., Beniston and Diaz, 2004). Long-lasting and severe heat waves in Alsace occurred in the following years: 2015 (100.9°C of cumulative heat stress over 27 days), 2003 (99.9°C of cumulative heat stress lasting over 19 days), 2006 (58.7°C of cumulative heat stress lasting over 22 days), 2017 (58.8°C of cumulative heat stress lasting over 17 days) and 1976 (43.1°C of cumulative heat stress lasting over 18 days). In the DE hop areas, both the longest heat wave duration and joint drought-heat waves (lasting 23/17 days) in 2018 were recorded. According to the cumulative heat stress index, these events had a lower magnitude than those that occurred in summer 2015, as also documented in Lhotka et al. (2018a); (Lhotka et al. (2018b). The extreme severity of the heat waves in 2015 was found in the majority of hop producing countries (e.g., DE, FR, CZ and SI) but the highest drought-heat stress occurred in Alsace, Tršice and Polabí hop sites (Figures S5,S6). Its severity (TS30 ≥ 72.2°C) was comparable with the drought-heat events that occurred in 2007 and 2012 (TS30 ≥ 70.1°C) in SI. In the ES hop regions, the most intense heat waves and drought-heat waves occurred in 2003 and were more than 1.5 times more severe than the heat waves in the UK, FR and SI. The drought-heat waves of 1994, 1986 and 1983 were among the most pronounced top 10 events in the DE, CZ and Poland hop regions, which is consistent with Tomczyk et al. (2019). Another intense drought-heat wave occurred in 2006 (Rebetez et al., 2009) when extremely hot and sunny weather caused very low values of alpha content in the CZ, DE, UK, FR, ES and SI.

3.3 Joint return periods of concurrent dry-cool, dry-hot, wet-cool and wet-hot modes

To determine the risk occurrences of four compound modes for each month of the hop-growing season, we applied the copula technique. Based on the marginal distributions of compound events and the parameters of the selected copula functions, we constructed a simple joint probability distribution for compound events. The joint return periods of concurrent dry-cool, dry-hot, wet-cool and wet-hot modes are summarized for each month of the hop-growing season for the entire study period (Figure S7). A comparison of the changes in the likelihood of such joint climate extremes was then conducted between the historical period (1961–1998) and present period (1999–2018; Table 3). Similar to the global warming signal, we found a substantial reduction in the likelihood of dry-cool and wet-cool events during the last 19 years, while widespread increases in the occurrence probabilities for the dry-hot mode are seen for most parts of the EU hop-growing areas. We determined that dry-hot events occurred 3–5 times per GS in the historical period, while in the present period, dry-hot events occurred more than 6–10 times in each GS. This implies that days with dry-heat stress are twice as likely in the present period (i.e., a 1-in-35 day event) compared with the historical period (i.e., a 1-in-75 day event). The chances of the occurrence of joint dry-hot conditions are generally higher than the chances of joint wet-hot extremes over the study region. Dry days are warming consistently more rapidly during April–June in the UK, April–July in PL/DE, April–August in FR, April–May in CZ, May–June in ES and June–August in SI. This implies that the high frequency of dry-hot events in the first part of the GS causes faster vegetative growth and earlier blooming and hop cone formation on the lower insertions of hop plants (Weihrauch et al., 2012). Related to this, the number of hop cones on the upper insertions decreased as well as the total yield because hop cones on the lower insertions cannot compensate for the number and size of cones on the upper plant insertions. At the beginning of June, the tallest plants began flowering in spite of the fact that they had not yet finished their vegetative growth. The result was early flowering after pruning, which agrees with other experimental studies (e.g., Donner et al., 2020). The basic feature of such irregular hop growth and hop cone development during dry-hot events is faster generative development in comparison with vegetative growth. The consequences are significantly lower yields and decreases in hop cone quality in comparison with normal years.

3.4 Hop yield losses related to the four compound extreme modes

Although evidence of hop yield loss due to drought is abundant, disentangling the roles of the tight coupling between heat and drought stress for determining yields has proved to be challenging and is largely unexplored. An outlook of the trends of hop yields, areas harvested and variabilities in the standardized yield residuals series for the seven studied hop-growing countries is summarized in Figure S8. The average hop cone yields ranged from 1.3 to 2.2 tha⁻¹ for DE, 0.9 to 2.2 tha⁻¹ for FR, 1.1 to 1.8 tha⁻¹ for the UK, 0.7 to 1.9 tha⁻¹ for PL, 0.9 to 1.9 tha⁻¹ for SI and 0.5 to 2.1 tha⁻¹ for ES. The CZ hop yield ranged from 1.1 to 1.6 tha⁻¹ with the highest losses recorded in 2015 and 2018 (down by more than 50%). The greatest losses were recorded in the Moravian Tršice region with 0.8 tha⁻¹ for the Saaz variety and 0.93 tha⁻¹ for all varieties.

We found different responses of hop yields to concurrent dry-hot conditions in contrast to heat stress-only during the growing season. Hop-cone formation decreased by 55.2% as the frequency of both dry and hot days increased (Figure S9). The years with the highest frequency of dry-hot days led to decreased yields of noble aromatic hops of 29–68% (forcing many farmers to harvest early). Correlation coefficients between hop yields and precipitation below the 35% quantile threshold (P35) and the yields with T_max above the 85% quantile threshold (T_max85) for the seven studied hop-growing countries are shown in Figure S10. Significant negative correlations were found for T_max85 from July to August and yields (−0.39 < r < −.70; p < .01) in DE, CZ, FR and SI. However, moderate frequencies of hot days during the same period had positive effects in the UK and PL. Extreme precipitation showed a slightly lower correlation than temperatures and the impact of P35 began to decline after June (hop plants reach their full flowering stage and their cones begin to form).

Figures 11a–c summarize the assessment of the degree of risk of lower yields of aromatic hop cones coupling the T_max and P anomalies over the EU. The degree of risk was estimated to be higher over extensive areas throughout the CZ and DE during the 1999–2018 farming years. However, 22.4, 12.5 and 7.2% of EU areas with conditions suitable for commercial hop production fell in the moderate, high and very high yield loss risk categories, respectively. Hop yields are currently most prone to dry-heat induced stresses and the concentration of hop cultivation in small EU regions makes these yields more vulnerable (Možny et al., 2009).

The statistical modelling risk estimated (Figure 11b) moderate to high yield losses in 2015, which is well in line with European reports of hop production (EU report, 2020). In 2018, the degree of risk was estimated as very high with extremely high intensity (Figure 11c), and concurrent extreme high temperatures and low precipitation caused low hop yields in the top producing regions. The statistical modelling risk matches the bimodal response of European hop yields to the observed climatic dipole of 2018 very well. That is, extreme dry-hot events were detected in north and central Europe with negative effects on hop yields, but relatively wet and cool conditions with positive impacts on yields were seen across large parts of the southern half of the study region. The observed yields under dry-hot conditions were greater than estimated, which means that the yields would have been much lower without irrigation. Irrigated hop yards are significantly more numerous in the southern half of the study region than elsewhere across the EU. The modelling risk suggests that the observed yields are strongly influenced by irrigation, which can alleviate the impacts of extreme temperatures. In this context, the European hot drought in 2018 can be considered as a key event for research on the vulnerability of hop production.

The risk occurrences of yield losses related to the four compound extreme modes (e.g., dry-hot, wet-hot, dry-cool and wet-cool) from bud burst to leaves (April) and from cone development to harvest (August) are summarized in Figure 12a–g. Generally, integrating the damage between April and August indicates that more than 62.7% of total yield losses are due to high temperatures under dry conditions and that 21.5% of total yield losses are due to dry-cool conditions in all hop-farming regions. For DE, the highest probability of yield losses was associated with the dry-hot and wet-cool modes, while the dry-cool mode was linked with moderate yield losses. The chances of occurrence of dry-hot conditions are 39.6% in May, 37.7% in June and 22.6% in August, which have led to six, five and four low-yielding years, respectively. There is a 28.3% chance of wet-cool conditions in April, which have led to four low-yielding years (Figure 12a). For CZ, the highest probability of yield losses was associated with dry-hot conditions in August (seven low-yielding years), May (six low-yielding years) and June (five low-yielding years), for which the chances of occurrence of dry-hot conditions were 27.6, 39.7, and 29.3%, respectively (Figure 12b). For the UK, dry conditions reduced yields less significantly at high and low temperatures than in DE and CZ; however, yield losses for the dry-hot mode in May–June (seven low-yielding years) prevailed (Figure 12c). For PL, both the dry-cool and wet-cool modes were linked to the highest yield losses from the first leaf formation to cone development. The highest probabilities of yield losses were associated with the wet-cool conditions of August (six low-yielding years) and dry-cool conditions of May (five low-yielding years) in which the chances of occurrence were 31.0 and 29.3%, respectively (Figure 12d). Thus, wet-cool conditions can also cause yield damage through waterlogging, increased pest numbers, increased fungal pathogens, and by loss of nitrogen fertilizer (Jursík et al., 2018). The yield losses due to drought that were further aggravated in the presence of high temperatures found in FR (Figure 12e) were rather large compared to the UK. Nevertheless, cold temperatures associated with dry conditions have led to moderate yield increments. For ES and SI (Figures 12f,g), the highest probability of yield losses also fell during dry conditions under high temperatures, particularly in April and August (each with six low-yielding years). The adverse effects of wet-cool conditions on hop yields in ES, as found in this study, seem counterintuitive at first glance, but this could be because hops are mostly irrigated in ES (Figure 12g). When summarizing the results among the major hop producers, it is notable that DE, FR and CZ appear to be more vulnerable to dry-heat stress for hop growing than the UK and Pl, whereas the dry-cool and wet-cool sensitivity of hops was comparable.

4.1 Based on the results obtained, the following conclusions can be highlighted

(a) Hop yields decreased by more than 28% in dry-cool conditions, while in dry-hot conditions, the yields dropped by 35–68% in the main hop-farming regions within the EU. Farmers have stabilized and increased the yields, but they have also spent more money on irrigation. This is because enhancing the yield and quality of hop cones means that spring/summer irrigation is essential for maintaining adequate soil moisture levels.

(b) It has been estimated that changes in the hop productivity in Spain over the last decades were strongly related to technology development and that the effects of dry-hot events were relatively moderate. For the UK and PL, hop yield losses associated with excessive precipitation had the same magnitudes as those associated with drought conditions; however, high temperatures in April–May can sometimes improve yields by increasing evaporative demand and reducing soil moisture to more optimal conditions. The yields in DE and CZ, the largest hop-producing countries, have decreased more than in any other country as the severity of CEs increases. This study also implies that the occurrence of days with heat stress and/or coupled drought-heat stress may occur anywhere in the hop-growing regions during the beginning of flowering to cone development, thus indicating that appropriate adaptation strategies are needed in all European hop-producing countries.

(iii) The coupling effects of CEs may gradually lead to changes in the regionalization of hop production. Policy assistance may be necessary for the adaptation of the EU hop-growing industry to changing climatic conditions. Even with the modest warming so far experienced, yields have stagnated and quality has declined. This fact means further expenses for premium beer production, which also contains aromatic hops.