Quantifying the Dunkelflaute: An analysis of variable renewable energy droughts in Europe

A cumulative frequency-duration distribution shows how often VRE droughts that lasted at least a certain duration occurred on average per year in the investigated data (Figure 1). This visualization is particularly useful to illustrate the occurrence of shorter-duration events, which are relevant for the operation of short-duration flexibility options in the power sector, e.g. battery storage. Note that the longest drought events, which may drive the need for long-duration electricity storage, are not visible here as the axis is truncated. The following sections focus more on such extreme periods. For a given threshold, the frequencies of all events that are at least as long as a given duration are shown in ascending order. Thresholds are scaled relative to the multi-annual mean of the respective time series (see Section 4).

In general, very short drought events occur much more often than longer ones for every given threshold. Events with very short duration reflect typical diurnal renewable variability, and should not be interpreted as exceptional periods of low availability. Long droughts lasting more than a week are infrequent. Importantly, the threshold choice substantially impacts the frequency of drought events. For lower thresholds, we generally find fewer events compared to higher thresholds. An exception is the left-hand side of the frequency-duration distributions shown in Figure 1. For very short drought durations, we find the largest number of events for relatively low thresholds, and a decreasing frequency of droughts for higher thresholds. This is because higher thresholds tend to identify fewer, yet longer-lasting droughts, which combine multiple shorter events that are counted as individual events for lower thresholds.

The cumulative frequency-duration distribution substantially differs between wind and solar power. For PV and low thresholds, we identify many events that last less than one day. They reflect the typical diurnal fluctuations of solar PV, with zero production at night-time [ueckerdt_analyzing_2015]. Increasing thresholds lead to fewer PV drought events, as many night-time periods of zero electricity generation merge into multi-day events. When the threshold approaches $\tau=1$, the identified droughts increasingly reflect a strong solar seasonality in Europe, with lower availability in winter than in summer. For on- and offshore wind power, droughts lasting less than one day occur less often. This is because wind power does not have regular non-availability at night-time as solar PV, but fluctuates less regularly. In turn, events that last longer than one day are more frequent, reflecting a typical multi-day variability of wind power [apt_spectrum_2007, ueckerdt_analyzing_2015]. The frequency-duration distribution of VRE portfolio droughts reveals a portfolio effect, i.e., droughts are less frequent when wind and solar PV are combined due to complementary wind and solar availability. Hardly any VRE portfolio droughts arise for lower thresholds. For higher thresholds, their frequency still is much lower than for single technologies (compare marked exemplary frequencies in each column in Figure 1).

Droughts occur less often in an ideal European interconnection than in single countries due to a geographical balancing effect. This is true for both single VRE technologies and portfolios. The reason for this is that periods of low solar and wind availability are not perfectly correlated across Europe [roth_geographical_2023]. For example, assuming a threshold $\tau=0.75$, there have been on average approximately eight, five, or four VRE portfolio droughts per year in Germany, Spain, or Europe that lasted at least one week, respectively.

In the SI, we provide additional results on the seasonality of drought patterns for Europe under the assumption of perfect interconnection. For low thresholds and events lasting longer than a few days, PV droughts are more frequent in winter, while wind droughts are more frequent in summer (Figure SI.3). The SI also shows return periods, which are the reciprocals of yearly frequencies. Similar to frequencies, return periods vary substantially between technologies and regions (see Section SI.3.)

Figure 2 shows the longest droughts obtained from the data sorted by year. A general finding is that maximum drought duration strongly depends on the underlying drought threshold. For very low thresholds around $\tau=0.1$, only very short droughts can be found. For high thresholds approaching $\tau=1$, the drought duration grows strongly and reaches a plateau of 365 days, i.e., a full year, in most weather years. This is because the search algorithm identifies very long events that reflect below-average yearly renewable availability at very high thresholds. Droughts that are relevant for energy system operations are likely to be found between these two extremes. Using very high thresholds close to $\tau=1$ thus appears not to be meaningful as these are likely to identify below-average weather years rather than drought events within a given year.

For medium thresholds, we find that both a technology portfolio and geographical balancing strongly decrease the maximum drought duration. For example, for a threshold of $\tau=0.75$, the longest PV drought in Germany lasts 291 days, while the maximum on- and offshore wind droughts are 303 and 209 days, respectively. In the case of a combined VRE portfolio, the respective maximum drought in Germany decreases to 106 days (portfolio effect). If, on top, perfect interconnection in Europe is considered, the duration of the longest event for the same threshold shrinks to only 55 days (balancing effect).

Especially for medium thresholds, the maximum yearly drought duration strongly depends on the weather year. For example, the longest VRE portfolio drought in Germany obtained for a threshold of $0.75$ occurred in 1996 (106 days), while the shortest yearly maximum drought duration was in 2018 (22 days). Considering perfect interconnection across all countries and the same threshold, the longest and shortest portfolio droughts occurred in 1997 and 1999, respectively (55 and 8 days). The most extreme droughts vary substantially across countries, both with respect to their duration and the year of occurrence (Figures SI.5 y SI.6). Importantly, the ranking of years varies with the drought threshold (Figure 3). For instance, the longest VRE portfolio drought in Germany for a threshold of 0.4 occurred in 2007 but for the thresholds 0.6 and 0.8 the most extreme droughts were in 1997 and 2003. The corresponding longest drought events in Spain occurred in 1991, 2017, and 2017. The spreads between the longest drought events in each year further increase with higher thresholds, before they decrease again approaching the threshold $\tau=1$ (Figure SI.7). From these findings follows that any claim on a specific year being particularly relevant for renewable drought analyses should always be qualified with the respective threshold.

The maximum drought duration varies substantially across seasons and technologies. Solar PV droughts primarily occur in fall and winter, with their median hour in November, December, or January (Figure 4). The maximum European PV drought under unconstrained geographical balancing lasts 248 days with a median hour in December for a threshold of $\tau=0.75$. The longest PV droughts with median hours in spring or summer months barely last longer than a few days, except for those identified by very high thresholds. In contrast, the longest European on- and offshore wind droughts mainly occur in spring and summer, with medians in July and maximum duration of 218 or 204 days, respectively. Maximum wind drought duration in winter months are much lower. In single countries, summer and winter wind droughts can be significantly longer (balancing effect). The seasonal complementarity of wind and solar further mitigates combined technology portfolio droughts to a substantial extent (portfolio effect). The longest-lasting droughts occur during winter months, which is a result of low seasonal PV availability in combination with rare but severe concurrent wind droughts. The median hours of the longest technology portfolio droughts in Germany (106 days) or Spain (137 days) occur in November, or, under the assumption of unconstrained balancing, in January (55 days).

Comparing weighted averages over all thresholds and countries, the maximum drought duration of a VRE portfolio decreases by 64%, 52%, or 47% compared to solar PV, onshore wind, or offshore wind alone (Table 1). The portfolio effect is more pronounced for PV than for wind power, as onshore and offshore wind power complement both diurnal and seasonal shortages of PV. In contrast, wind power does not incur diurnal shortages. Assuming a European copperplate (CP), the portfolio effect (-80%) for solar PV is even stronger than for any isolated country. This is because in a larger geographical area, there is more complementary wind power available to compensate for solar PV shortages [roth_geographical_2023]. In turn, the portfolio effect is less pronounced for onshore wind power in the European interconnection (-45%) as PV availability is more homogeneously distributed. For offshore wind power, the portfolio effect is higher (-70%), as only a subset of countries features this technology, such that solar and wind onshore resources of all countries can compensate for extreme offshore droughts in a subset of countries.

Considering unconstrained geographical balancing across all countries, the longest PV, onshore wind, or offshore wind droughts shorten by 1%, 46%, or 34%, using weighted averages over all countries. The very small average balancing effect for PV, however, conceals that there is a substantial north-south divide between countries with higher and lower solar irradiation. For South European countries with abundant solar availability such as Spain, Portugal, Italy, or Greece, geographical balancing even increases solar PV drought duration. In turn, geographical balancing alleviates maximum PV drought durations in North European countries such as Germany or Scandinavia. In contrast, the balancing effect on wind drought duration is more similarly distributed across countries, as maximum wind droughts rarely occur simultaneously throughout Europe. The balancing effect is even more pronounced for VRE technology portfolios, decreasing by 65% on average. That is, geographical balancing not only helps to mitigate droughts for single renewable technologies, but even more so for a technology portfolio, due to complementary technology portfolios with different capacity mixes across countries.

The results discussed so far reveal a strong sensitivity of drought characteristics to the underlying threshold used for drought identification. Focusing on events lasting at least two days, Figure 5 exemplarily illustrates drought patterns for a wide range of thresholds for the years 1996 and 1997. In general, single-threshold analyses using low thresholds can detect extreme short-duration droughts, which may occur in isolation or adjacent to high-availability periods (white spaces in the Figure). Higher thresholds increasingly identify longer-lasting events. These potentially include consecutive shorter periods of extreme renewable shortage or surplus, which are smoothed by the averaging mechanics of the identification algorithm (see Section 4.1). The multi-threshold perspective illustrated in Figure 5 demonstrates that longer-lasting events identified by higher thresholds may encompass multiple severe droughts detected at lower thresholds.

Solar PV events identified by moderate or high thresholds can last several months, indicating low solar availability during winter in Europe (Figure 5, panel b). Lower thresholds identify very severe PV droughts driven by low seasonal availability in combination with factors such as persistent cloud or snow coverage, such as in winter 1996. In summer, only rare and isolated PV droughts of lower severity occur. These seasonal PV characteristics can generally be found across all years and are often more pronounced in Northern Europe, e.g., observable in winter 2011/12 (compare Germany or United Kingdom to Spain in Figure SI.9). In contrast, severe long-lasting onshore and offshore wind droughts can occur throughout the year but tend to be more frequent in summer (Figure 5, panels c and d), which is in line with general wind seasonality [kaspar_climatological_2019]. The multi-threshold perspective further reveals that severe wind droughts identified by lower thresholds can be sequential and may occur within contiguous below-average wind periods detected at higher thresholds. The latter may last up to several months. While such long-lasting events usually occur in summer (compare Figures SI.9-SI.11), they may also take place in winter in some countries (e.g., in 1996/97, Figure 5). On- and offshore droughts are generally correlated, but may differ in severity (Figure 5, panels c and d).

The multi-threshold illustration also shows that wind and solar power droughts are often complementary. Typically, high solar availability offsets wind droughts in summer, while wind power compensates for the low solar PV availability in winter. This results not only in briefer but also less severe droughts, i.e., higher thresholds apply in the same periods compared to single technologies (portfolio effect, compare panel a of Figure 5 with panels b-d). However, if severe and persistent wind and PV droughts coincide, pronounced long-lasting portfolio droughts can occur, as seen for instance in winter 1996/97. Geographical balancing can mitigate the severity and duration of such compound events as solar seasonality is lower in Southern Europe and wind droughts do not occur simultaneously across countries (balancing effect). Severe and long-lasting portfolio droughts may arise in single countries and also across Europe at the beginning of a calendar year, (e.g., observable in 2012, Figure SI.9), at the end of a year (e.g., 1982, Figure SI.11), or span across the turn of years (e.g., in winter 1996/97, Figure 5). In some years, winter portfolio droughts are much less pronounced (e.g., 2013/14, Figure SI.10). Countries relying heavily on wind power, such as the United Kingdom, may also experience very severe events in summer.

We introduce a novel “drought mass” metric to identify the most extreme portfolio drought events and show that it identifies events that are relevant for long-duration electricity storage in renewable European energy systems. The drought mass rates both the duration and severity of contiguous events by accumulating and ranking scores for all droughts identified at thresholds $\tau\leq 0.75$. Section 4.2 further elaborates on the drought mass mechanics.

Figure 6 shows extreme portfolio drought patterns identified by the drought mass metric for the years 1996/97. The most extreme events, i.e., those with the highest drought mass, comprise sequences of shorter, but more severe droughts within contiguous well-below-average periods that may last up to several months. Events with the highest drought mass scores may occur in winter (teal boxes), potentially spanning across the turn of a calendar year, as in the case of the European copperplate in 1996/97. Severe drought mass events may also occur in summer (gray boxes) in countries that heavily rely on wind power such as the United Kingdom or Poland.

Note that renewable availability is not zero during these events, but it is low on average for a very prolonged period of time. For instance, the average availability factor of a renewable technology portfolio during the most extreme drought is 0.11 for the European copperplate scenario (55 days in winter of 1996/97), 0.07 in Germany (109 days in winter 1995/96), and 0.08 in Spain (131 days in winter 1988/89). For comparison, the average portfolio renewable availability factor over all hours in the data is 0.23 for the European copperplate, 0.19 for Germany, and 0.21 for Spain, respectively. That is, average renewable availability still amounts to 47% of its long-run average during the most extreme drought event that defines long-duration storage needs in a fully renewable, perfectly interconnected European power sector. For Germany or Spain, the average availability during the most extreme drought amounts to 37% or 35%, respectively. Yet, within the largest drought mass events, we find more severe but shorter drought events with very low VRE availability. For instance, during the extreme drought in the winter of 1996/97, events that lasted 17 and 18 days occurred in Germany with average availabilities of just over 0.05 identified by a threshold $\tau=0.5$, relating to 27% and 29% of the long-run average, respectively.

Peak electricity demand often depends on ambient temperature and occurs in winter in most European countries, except for the Mediterranean area. Accordingly, winter drought events that coincide with high electricity demand are particularly relevant. Using a stylized electricity sector model (Section 4.3), we find that these compound events are a major driver for the use of long-duration electricity storage in all three interconnection states. Figure 6 shows the optimal operation of long-duration electricity storage for different degrees of interconnection between countries as well as electricity demand patterns for the years 1996/97. The most severe identified droughts largely coincide with the periods of long-duration storage discharge, i.e., decreasing storage levels. This holds true for both the European copperplate scenario and several isolated countries in the island scenario. When considering policy-relevant interconnection across all countries (“TYNDP”), the storage level patterns do not structurally change but only show a level shift compared to the island scenarios. That is, drought patterns identified for isolated countries are a reasonable approximation of drought patterns in settings with more policy-relevant interconnection with regard to long-duration storage operation.

In some countries, the most severe drought mass event does not coincide with the major discharging period of long-duration storage. For example, in Sweden, Poland, or the United Kingdom, the metric finds that the largest drought events occur in summer (gray boxes in the lowest three panels of Figure 6). Long-duration storage usage, however, still coincides with the most extreme winter droughts, which are comparatively less severe. This is because electricity demand is lower during the more severe summer drought events identified in these countries. Consequently, summer droughts do not affect the system’s ability to meet demand as much. In complementary model runs where we assume a flat electricity demand profile, i.e., abstracting from seasonal and diurnal variability, the most severe event measured by the drought mass metric perfectly coincides with storage-defining periods in all countries, including Sweden, Poland, and the United Kingdom (Figure SI.12).

The duration and severity of most extreme winter droughts captured by the drought mass metric varies significantly across years and countries (Figure 7). Assuming perfect interconnection between countries, the most extreme event in the data occurred in the winter of 1996/97 and lasted 55 days. This European super drought was caused by pronounced and partially overlapping droughts in several Central European countries and the United Kingdom (Figure 7). As the most pronounced events do not occur simultaneously in all countries, geographical balancing mitigates also such extreme droughts. Accordingly, the European super drought is substantially shorter than the most extreme droughts in nearly all isolated countries. Applying the drought mass metric to individual countries, we find the longest winter events in Eastern and Southern Europe. Further, smaller countries such as Slovenia (201 days, 1985/86) or Slovakia (195 days, 2015/16) tend to have longer extreme droughts than larger countries such as France (64 days, 2004/05), Sweden (83 days, 1997/98), Germany (109 days, 1995/96), or Spain (131 days, 1988/89). This is because smaller countries benefit less from geographical balancing within their borders.

Figure 8 shows four snapshots of selected hours within the European portfolio super drought, as identified by the drought mass metric at the turn of the years 1996/97. Each panel shows the drought severity present in the displayed hour for isolated countries (island scenario, left map) and for the European copperplate scenario (right map), illustrated by the lowest applying threshold. While the super drought affects multiple European countries simultaneously, the severity varies across countries and over time. Comparing the island scenario and the European copperplate shows that renewable energy shortages in the latter are always less severe than in the most affected countries at the same hour. This is because the assumed European interconnection mitigates shortages in one country by leveraging higher renewable availability in others. That is, geographical balancing is feasible even within the most severe pan-European drought event in the data. The graphs further illustrate that the super drought comprises sequences of shorter, but more severe droughts, both in individual countries (different colors on maps on the left-hand side) and on the European level (different colors between maps on the right-hand side). Further, even within the European super drought, there are hours with relatively high renewable availability (Figure 8d). The progression of sequentially occurring events inside the European super drought is even more visible in a complementary animated graph which is available as supplementary information [kittel_high_2024]. The animation, covering December 1996 and January 1997, shows that very severe droughts that cover most of Europe (e.g., visible in Figure 8a) are relatively brief and occur rarely, even inside the super drought event. That is, some renewable generation potential is always available somewhere in Europe. Even during the most extreme droughts, this can balance low renewable availability across technologies and regions to some extent.

Our VRE drought analysis is based on VRE availability factor time series. We use a new open-source method that applies a variable-duration mean below threshold (VMBT) algorithm for drought identification [kittel_measuring_2024]. This method searches for periods with a moving average below a given drought qualification threshold by iteratively decreasing the event duration. In each iteration, the algorithm sets the averaging interval to the respective event duration, starting with very long-lasting events and iteratively continuing to such that last only a few hours. A moving average below the drought threshold identifies a drought event. It is excluded from subsequent iterations, in which the averaging interval decreases further and additional (shorter) events are identified.

This iterative procedure overcomes shortcomings of previous research [kittel_measuring_2024]. It pools adjacent periods that independently may not qualify as VRE drought, identifies unique events, avoids double counting as well as overlaps with adjacent events, and ensures that the full temporal extent of drought periods is captured [kittel_measuring_2024]. To cover the full spectrum of VRE droughts, we parameterize the VMBT algorithm such that it searches for periods ranging in duration from two years to one hour. The code and input data of the drought analysis tool are freely available on GitLab at https://gitlab.com/diw-evu/variable_renewable_energy_droughts_analyzer.

To account for varying renewable generation potentials in the different settings, i.e., single technologies or different VRE portfolios across countries, we use a large number of drought thresholds and scale these relative to the multi-annual mean availability of each time series under investigation [kittel_measuring_2024]. Thresholds range from 10% to 100% of the mean availability factor across all investigated years, increasing in 5% increments. Very low thresholds around $\tau=0.1$ identify very severe droughts with near-zero VRE availability. In contrast, very high thresholds near $\tau=1.0$ are likely to identify below-average weather years. Figure SI.1 illustrates how these relative thresholds relate to absolute capacity factors.

We use country-level VRE availability time series from the Pan-European Climate Database provided by the European Network of Transmission System Operators for Electricity [de_felice_entso-e_2022], comprising 38 weather years from 1982 to 2019. These data have been used for policy-relevant strategic reports, such as the Ten Year Network Development Plan (TYNDP) 2022 or the European Resource Adequacy Assessment 2021. The renewable capacity assumptions used to generate the capacity-weighted VRE portfolio time series stem from the TYNDP 2022 (scenario Distributed Energy). For Germany, we update the capacity targets according to the latest government targets [bundesamt_fur_justiz_gesetz_2024].

Multi-regional VRE drought analyses based on VRE availability time series can be conducted based on two extreme assumptions of electricity transmission between countries [kittel_measuring_2024]: either perfect interconnection across all countries (“copper plate scenario”) or complete isolation of all countries (“island scenario”). For the former, all countries are treated as one single pan-European node. All regional time series are combined into a composite using capacity-weighted averages, with weights according to the capacity assumptions from TYNDP 2022. For the latter scenario, thresholds are scaled as discussed above. Accordingly, regional VRE portfolios are constructed using capacity-weighted averages of all contributing technologies.

We introduce a drought mass metric to identify storage-defining drought events. For each investigated threshold $\tau\leq 0.75$, we use time series that include the hourly availability factors of identified events, assigning a value of $0$ to hours that do not qualify as droughts according to a given threshold. We equally weigh across all thresholds by replacing the availability factors of identified events with the value $1$. For each hour, we then accumulate the resulting values. Within a given year-pair, we search for the highest cumulative multi-threshold score of contiguous events, using a cut-off threshold $\tau=0.75$. This approach defines the event duration according to the threshold $\tau=0.75$, while all thresholds $\tau\leq 0.75$ equally contribute to an event’s overall score. We tested a wide range of cut-offs and weighting schemes to identify the most effective drought mass design. The best alignment between events with the highest winter drought mass scores and major storage discharge events was achieved with a cut-off threshold of $\tau=0.75$ and equal weighting across all thresholds.

Countries with high shares of wind power in their capacity mix may experience the most extreme portfolio droughts in summer. Peak electricity demand periods often occur in winter, except in some South-European countries. To account for this, we compute a drought mass score for droughts occurring throughout the year and another one relating only to winter droughts excluding the period from May until September. When illustrating the relation of drought patterns and long-duration electricity storage use, we display both the most extreme summer and winter droughts if the highest drought mass score relates to summer droughts (compare regions with gray and teal boxes in Figure 6). Conversely, if the highest drought mass score throughout the year relates to a winter drought, we mark only one event (compare regions with a gray box only in Figure 6).

We use the open-source power sector model Dispatch and Investment Evaluation Tool with Endogenous Renewables (DIETER) to analyze the interaction between VRE droughts and long-duration storage needs in a fully renewable European power sector. DIETER is a linear optimization model that determines least-cost capacity and dispatch decisions based on an hourly resolution [zerrahn2017, gaete-morales_dieterpy_2021]. Different versions of the model have been used to study various aspects of VRE integration and their interaction with other flexibility options or sector coupling technologies [zerrahn2018, schill_long-run_2018, schill_electricity_2020, schill_flexible_2020, kittel_renewable_2022, roth_geographical_2023, kirchem_power_2023, gong_bidirectional_2023, gaete-morales_power_2024]. Here, we use the most recent extension of the model that represents 33 European countries (EU27, the United Kingdom, Norway, Switzerland, and the Western Balkans). As very long-lasting VRE droughts may span across the turn of a calendar year, we extend DIETER’s planning horizon to two full years. For transparency and reproducibility, we provide the model code, the input data, and a manual in a public repository under permissive licenses available on GitLab at https://gitlab.com/diw-evu/projects/quantifying_the_dunkelflaute_energy_system_analysis.

We use policy-relevant renewable capacity mixes from the TYNDP 2022 (scenario “Distributed Energy”). Electricity demand profiles are derived from the European Resource Adequacy Assessment 2021 (target year 2025), including limited electrification of heat and transport. These profiles are scaled to the TYNDP demand levels in 2050, and adjusted for net-importing and net-exporting countries to ensure they can be met by domestic renewable supply. Inter-annual variations are mainly driven by temperature differences.

The model features green hydrogen technologies, covering its generation, storage, and reconversion to electricity. We assume that hydrogen cavern storage can be expanded without restrictions in every country. This is a deliberate simplification, as we aim to illustrate the relation between VRE portfolio droughts and long-duration electricity storage, but not to derive policy-relevant geographical allocations of hydrogen storage across Europe [caglayan_technical_2020, talukdar_techno-economic_2024]. We focus on hydrogen used for long-duration electricity storage and abstract from additional hydrogen demand of other sectors, hydrogen imports from other world regions, and hydrogen exchange between countries.

We investigate three scenarios with varying degrees of interconnection between the 33 countries. First, we model an island scenario without any cross-country exchange of electricity, i.e., domestic demand needs to be satisfied by domestic supply only. Second, we allow for unlimited exchange of electricity across countries, which represents the European copperplate scenario. Third, we complement these extreme scenarios by allowing for a more policy-relevant cross-border exchange of electricity according to the Ten-Year-Network-Development-Plan 2022 (scenario “Distributed Energy”) to prove the validity of the drought mass metric as a proxy for very severe VRE droughts.

Figure SI.1 illustrates how the relative thresholds used in the search algorithm relate to absolute availability factors for each VRE technology and portfolio. Absolute thresholds are generally lower for solar PV than for wind power because of the lower average availability of solar PV.

Figure SI.2 illustrates the cumulative frequency-duration distributions for droughts lasting up to a full year for Germany, Spain, and the European copperplate scenario. Only higher thresholds identify droughts lasting longer than a few weeks. For thresholds $\tau<1$, portfolio droughts are generally briefer than single-technology droughts. To raise complementary insights on seasonality, Figure SI.3 shows seasonally differentiated distributions for the copperplate scenario. In general, longer droughts are more frequent in winter than in summer. Brief PV droughts are more frequent in summer and are detected by higher thresholds.

Return periods of VRE droughts are given by the reciprocal of yearly frequencies. The return period-duration distributions represent the right-hand side of the event distribution and indicate the longest event duration that can be expected to reoccur after a given number of years (Figure SI.4). Note that the return period monotonically increases and the maximum return period is limited by the 38 years of data that we investigate.

For each threshold, the maximum event duration increases with higher return periods (compare increasing contour lines and marked data points in Figure SI.4). Additionally, the return period-duration distributions are highly sensitive to the underlying threshold. For any given return period, higher thresholds lead to events with longer maximum durations, plateauing at 365 days for very high thresholds. Note that such high durations indicate below-average wind and solar years and should not be mistaken for actual drought events. The return period-duration distributions vary significantly between technologies and countries. For example, the maximum offshore wind drought for a threshold of $\tau=0.75$ in Germany that returns every 20 years lasts 198 days and is much shorter than the corresponding PV (289 days) and onshore wind droughts (297 days). By comparison, in Spain, such 20-year return period droughts are shorter for PV (216 days) and onshore wind (245 days) but longer for offshore wind (211 days).

When combined in a technology portfolio, the maximum duration decreases substantially for each return period in both Germany and Spain and also in the European copperplate, which is another instance of the portfolio effect described above. The balancing effect is also visible: with perfect interconnection, the maximum drought duration further decreases due to a limited temporal correlation of wind droughts across Europe. In the case of solar PV, higher solar availability in South Europe during winter balances lower availability in Northern Europe.

Figures SI.5 and SI.6 show the maximum duration of single drought events across all years for each country and threshold. Figure SI.7 illustrates the inter-annual variability of the maximum drought duration for each threshold and technology (portfolio).

Figure SI.8 illustrates identified droughts lasting longer than one week or more, filtering out briefer events. This highlights very long-lasting below-average wind periods that may encompass multiple drought events of varying severity and solar seasonality. Combined in a portfolio, these periods are less severe (portfolio effect) and are further mitigated when assuming unconstrained geographical balancing (balancing effect). Figures SI.9, SI.10, and SI.11 show identified drought patterns lasting at least one day for the years 1982/83, 2011/12, and 2013/14. Figure SI.12 additionally illustrates the most extreme drought events and optimal storage use in 1996/97 for selected regions assuming flat demand profiles, i.e., eliminating any diurnal or seasonal demand variability.