Wind and solar energy droughts: Potential impacts on energy system dynamics and research needs Open Access

As economies regionally and globally move toward increased dependence on wind and solar energy sources due to their low cost of energy and very low lifecycle carbon dioxide emissions, they must adapt their electrical grids to be robust to extremes of varying generation and demand, and in particular, to wind and solar energy droughts (WSDs). On the most basic level, a WSD is simply a period of time over which much less energy than normal is produced due to weather variability. The minimum levels of wind and solar energy production reached over large geographic areas during droughts can be significant. As an example, WSDs were investigated using a multi-decadal time series of a hypothetical mix of wind and solar generation evenly distributed across New York State, in which each provides 50% of the annual total generation.¹ In that study, electricity generation levels during the most extreme 1-, 10-, and 30-day WSDs were found to be approximately only 12%, 35%, and 60%, respectively, of the multi-year annually averaged power generation. Another example is the summer 2021 wind drought experienced over Europe, which was the lowest wind summer in the previous 60 years. During a period stretching through summer and early autumn, a UK energy company reported that renewable assets produced only 68% of the amount expected for that time of the year.² The minimum levels of renewable energy produced that are of importance to the energy system will vary regionally and over multiple timescales and have dependencies on short-term weather (the normal movement and evolution of high- and low-pressure systems), seasonal cycles that also vary regionally, and on multi-year to multi-decadal variations, including potential variations associated with climate change.

For most grid systems with small amounts of variable wind and solar generation, the greatest system stress occurs during periods of peak load, and measures taken to ensure system reliability are determined by those peak loads. As the amount of variable wind and solar generation on the grid increases, the greatest stresses and mitigation mechanisms will be increasingly co-determined by shortfalls of generation during WSD events and their alignment with periods of high loads (energy demand) on the system. If the grid has little storage capacity, the intensity of those droughts determines the amount of dispatchable generation capacity needed to ensure generation will always be able to meet demand (dispatchable generation units are those that commit with near certainty of being able to generate at a specific level at a future time). Also, as wind and solar capacity grows, being able to forecast periods of low renewable generation will become increasingly important, first to reduce costs associated with bringing those dispatchable generators online, second to reduce the possibility that insufficient dispatchable generators are available, and third, to proactively manage storage state of charge. As wind and solar generation increases further, potentially becoming dominant energy sources, addressing the impacts of WSDs transitions from principally being a forecasting problem to also being a grid planning issue. Knowledge of the characteristics of WSDs then becomes essential for determining the amount of new energy transmission, storage, demand response programs (voluntary reductions of energy demand during periods of high system stress), and additional generation capacity (variable and dispatchable) that need to be built to accommodate energy droughts. If WSDs are not planned for properly, they could potentially affect the reliability of the electric grid. Similar concerns for periods of extended high electrical load, in combination with weather that creates challenging operating conditions for generation facilities, have prompted regulatory action to maintain adequate supply in these conditions.³ 

Independent of reliability issues, even for grid systems with small amounts of wind and solar generation, droughts can be important from a purely economic perspective if they are intense and prolonged. An example of this is the wind drought that occurred across many parts of the United States during the first half of 2015. At the time, this resulted in serious concerns about the economic viability of those wind plants⁴ as it was unknown how long it would take for wind generation to return to more normal values, or how frequently droughts of similar intensity might occur in the future.

Just as renewable generation technology has been evolving rapidly over the past few decades, many forms of energy storage are being deployed and explored, all in the context of the development of technology for electrification of industrial processes. Electrification of transportation has created large markets for batteries that have resulted in falling battery prices and expanded the time-domain over which batteries are now the most economical form of storage. Parallel efforts focused on fertilizer, steel, and cement production are creating expectations of large markets for renewably generated hydrogen and heat. This may result in a very large demand for storage of hydrogen and heat for industrial purposes that could also be used to generate electricity during WSD periods. Because of the speed of change, and multiple technologies being deployed for generation and storage, when WSDs are analyzed, the metrics and modeling tools should be flexible so that results will be applicable in complex systems of weather-dependent energy generation, managed demand, and multiple means of electrical storage that will coexist in the future electrical grid.

There are four main attributes of both energy supply and balance droughts: intensity, duration, spatial extent, and frequency of occurrence. As noted by Raynaud et al.⁵ wind and solar energy droughts can be differentiated into energy supply droughts that consider energy generation only, or energy balance droughts that include the co-occurrence of supply droughts with large energy demand. Energy supply droughts occur when wind or solar energy generation are small compared to their climatologically expected values; energy balance droughts occur when generation minus load is small compared to its climatologically expected value for a given energy system. Like wind and solar generation, variability in energy demand is principally dependent on meteorological conditions, mainly temperature, and has strong seasonal variations; consequently, wind and solar energy supply droughts and energy balance droughts are both predominantly determined by weather conditions. Although the sensitivity of energy demand to weather, and temperature in particular, will change in uncertain ways with increasing electrification of industry and transportation and as technology evolves, weather will undoubtably continue to be the principal source of load variability.

Wind and solar supply droughts can be defined either relative to the seasonally varying mean production, or relative to the long-term, multi-decadal mean production. The former removes the mean seasonal variation leaving only short-term weather and interannual climate variations, while the latter includes the seasonal variations. Although both can have value in terms of understanding the meteorology associated with drought events, including the seasonal variations can be more useful if one wishes to analyze the impact of droughts on the energy system, as the system must be responsive to the total energy generation shortage, no matter what its source. This is especially true when considering energy balance droughts, since demand also exhibits seasonality. Numerous studies have demonstrated that for most locations wind and solar are complementary, so that to some extent the seasonal variation of energy supply or energy balance droughts can be reduced when combining wind and solar generation (for a review, see Wu and West⁶ and Jurasz et al.⁷).

Wind and solar droughts are often discussed in the context of a potentially realizable electrical grid in which generation resources, including wind and solar, are always sufficient to meet energy demand, even during WSDs. The remedy to episodes of low wind and solar generation during WSDs is some combination of other generation types (geothermal, biomass, nuclear, and fossil fuel); construction of wind and solar capacity larger than some defined baseline (overbuilding); additional transmission capacity beyond some assumed default level to allow averaging of demand and wind and solar supply over a larger area; multi-day energy storage to allow averaging of demand and wind and solar supply over longer periods; and demand management (that reduces demand during low wind and solar generation periods). As an aspirational goal, metrics for WSDs ideally should be designed to be easily related to the magnitude of storage, transmission, and overbuilding that would resolve the challenge of meeting demand at all times, including during WSDs, while also accounting for generation sources other than wind and solar, that may or may not be dispatchable.

Many different metrics have been proposed for WSDs. Following the nomenclature of Patlakas et al.⁸ and as discussed by Wilczak et al.,¹ most of these can be divided into two categories, Duration Given Intensity (DGI) or Intensity Given Duration (IGD). In the DGI approach, thresholds of intensity are first selected, and then droughts of variable duration are defined as those periods when the intensity exceeds that threshold. In contrast, with the IGD approach a fixed drought duration is selected, and then intensities of the worst drought(s) are found for that duration. The most serious shortcoming of the DGI and IGD approaches is that they depend on arbitrary thresholds, such as percentiles of a distribution, that are independent of the energy system itself. Kittel and Schill⁹ described some of the challenges that occur when using threshold-dependent drought metrics.

In contrast, the running mean-based, return period method of Wilczak et al.¹ and also identified by Duchene et al.¹⁰ defines droughts for a given region based on their return period (the statistically expected time between a defined set of events), and then using continuous moving averages (durations) of drought intensity, implicitly finds the corresponding intensity threshold needed for that return period. Energy balance droughts defined in this way, dependent on their return period, would then be prescribed in terms of an attribute of the energy system, that is, the tolerance of the grid (or society) to accept these shortages at some frequency of occurrence, and mitigate them through a combination of energy storage, demand response, dispatchable generation, or load shedding. For energy supply droughts, application of this method provides the return period of supply shortfalls, and these can be analyzed as a function of load (or temperature as a proxy for load) to provide insights into the joint distribution of supply shortfalls and demand.

Wind energy droughts usually occur when persistent high surface pressure is present over a large region, often associated with blocking highs. For solar energy, droughts occur during protracted periods of cloud cover, which can be associated with low-pressure systems, or with fog and low clouds that occur in some high-pressure regimes. Compound energy supply cases combine weak winds with cloudy days, and the exact meteorological conditions of most importance will depend on the ratio of installed wind and solar generation capacity. Energy balance droughts that combine weak winds, cloudy days, and either hot or cold temperatures may stress the system the most,¹¹ and because of the complexity of the meteorological state and reduced sample size of these compound events, it may be more difficult to classify them into simple regimes.

Some work has been done to classify meteorological conditions present during drought events in Europe,^12–15 relating wind droughts for example to Scandinavian blocking highs. Overall, however, this is an area where much more could be done in other regions, especially given the likely significant regional variability in meteorological drivers of energy droughts.

Ice accumulation on wind turbine blades also can result in significant power reductions, of up to 80% even in high wind conditions,¹⁶ and can cover large geographic areas at times when energy demand is high, leading to a wind energy drought event. High wind speeds above wind turbine cutout levels likewise can produce large drops in wind energy production across affected areas. However, these impacts are typically short-lived and usually coincident with higher levels of generation in nearby regions. For this reason, it is less likely that significant wind energy droughts will be caused by high wind speed events.

Snow and ice on PV panels will also greatly reduce solar generation, and rooftop solar may be especially prone to long periods of low generation from snow. Although potentially a major issue, in part because of the co-occurrence of snow and high energy demand during cold temperatures, the authors are not aware of any studies on the impact of snow on solar energy droughts. The magnitude of droughts caused by snow on rooftop solar PV panels may require the use of the Observed Power Generation (OPG) method (see below), as it is unlikely that sufficient information exists on individual panel tilts and orientations to calculate these effects using a bottom-up approach. Also needed will be an understanding of the efficacy of snow and ice remediation efforts for utility-scale solar installations. In addition, high aerosol concentrations from wildfires,^17,18 dust storms,¹⁹ or volcanic eruptions can all contribute to reduced solar energy generation and potentially contribute to extreme drought events.

As hydrologic droughts generally are more widely or intuitively understood than WSDs, it can be useful to compare and contrast the two. One key difference between WSDs vs hydrologic droughts is that ground water, soil moisture, and water stored within plants all represent natural hydrological storage reservoirs. In contrast, wind and solar energy have no such natural storage.

Because of natural storage in soil moisture, the most severe impacts of hydrologic droughts occur with a considerable time lag, months to years, from the initial lack of precipitation. Knowledge of the slow development of hydrologic droughts through drought monitoring efforts,²⁰ therefore, allows organizations to act and prepare responses to those droughts. Because of the lack of natural storage of previous energy generation, WSDs in energy systems without long-term storage will have a near immediate effect, and energy droughts can be impactful on daily or even shorter timescales. Inclusion of energy system multi-day to seasonal storage capability mimics the role of soil moisture or snow pack for hydrologic droughts. Monitoring of energy droughts will therefore be simpler than for hydrologic droughts, because with no natural storage, only the energy state of battery, hydrogen, or other designed storage systems needs to be known.

For these reasons, the development of observing networks, associated infrastructure, and tools for energy drought monitoring is less compelling than for hydrologic droughts. A more compelling use of knowledge of energy droughts is to use that understanding to rationally and optimally develop the infrastructure for a future highly wind- and solar power-based energy system. A thorough understanding of these droughts can provide confidence that design decisions made for such a system are robust to extreme events, and how different components of that energy system will function in the presence of these extreme events.

Another difference between WSDs and hydrologic droughts is that hydrologic droughts spanning a wide range of intensities can have serious societal and economic consequences as crop-yields decrease. Those consequences occur over a continuum of conditions, with no clear thresholds. For a highly renewables-based energy system with significant energy storage and demand response capacity, costs to build a system resilient to droughts also will increase continuously with the magnitude of the droughts expected, but once built, no new costs would occur. However, if a drought of larger magnitude than planned for occurs, the energy system would function smoothly until the point that storage is depleted, demand response no longer responds, and load shedding occurs. Thus, a threshold exists on the impact of wind and solar droughts on electric grid stability, with no similar analogy for hydrologic droughts: either there is sufficient power to avoid load shedding or there is not. Crossing that threshold can mean having a sudden jump in the severity of its impact. Although caused largely by fossil fuel infrastructure failures, an example of significant load shedding occurred during winter storm Uri (13–17 February 2021) which caused rolling power outages in parts of Texas. There were at least 57 deaths associated with that storm and the resulting outages, with damage estimates exceeding $195 × 10⁹,²¹ making it one the worst weather disasters in Texas history. This illustrates the importance of designing an electric grid system so that it can accommodate the worst-case energy drought scenario.

Because of the multi-decadal time spans they cover and the physical consistency of their wind, cloud, and temperature fields, reanalysis datasets have been the most common data source used for wind and solar energy drought analyses. Since they assimilate observations, reanalyzes are often considered to represent the true state of the atmosphere when evaluating wind and solar energy generation. However, since reanalyzes are themselves dependent on an NWP model, they too have errors, and comparisons of reanalyzes to wind and solar insolation observations often show those errors can be significant.^22–26 This is especially true for extreme WSDs, where for example Wilczak et al.²⁷ showed that when averaged over observation sites that span the United States, the ERA5-derived solar power produced during the most extreme 10-day duration droughts was 62% greater than that derived directly from solar irradiance observations; for the most extreme wind droughts at some sites even in flat terrain, the ERA5-derived wind power was less than half of that observed. This indicates that any drought analysis based on the uncorrected ERA5 could have significant limitations.

Another limitation of coarse-resolution reanalyzes is that they may not properly represent topographically forced flows in regions of complex terrain. These sub-grid scale topographic flows may affect the ability of coarse resolution reanalyzes to properly characterize wind drought events and lead to a local overestimation of wind drought amplitudes.

We also note that some wind drought studies, especially those using climate models, have relied on 10-m surface wind speeds extrapolated to hub-height (∼100 m), usually because the model did not output winds at hub-height. If the wind speed at hub-height is low, it is likely that it is also low at 10 m. However, the reverse is not always true, as during periods with strong stable stratification, winds near the surface can be very weak, but winds at hub-height, perhaps associated with low level jets, can be strong enough to produce significant wind power. Extrapolation of surface winds to hub-height could then result in an overestimate of the amplitude of wind drought events. More generally, due to the diurnal variation of boundary layer turbulent mixing, on average 10-m surface winds are weak at night and strong during the day, while turbine hub-height winds are strongest at night and weaker during the day.

For solar droughts, an alternative to NWP model-based reanalyzes is to use those from radiative-transfer models that are driven only with satellite cloud observations. There are two such products currently available: the National Solar Radiation Data Base (NSRBD),²⁸ which predominantly covers North and South America; and the Surface Solar Radiation Data Set—Heliosat (SARAH),²⁹ which covers Europe, Africa, and parts of the Middle East and South America. Because these two datasets are based on visible and infrared cloud imaging products, which are not assimilated into standard NWP-based reanalyzes (other than for wind speed estimation through tracking of cloud elements between sequential cloud images), one would expect that solar droughts would be characterized more accurately than in NWP-based reanalyzes. However, to our knowledge, there are to date no published analyses of the accuracy of the NSRDB or SARAH products for solar drought events, nor detailed analyses of solar drought characteristics using these products.

A limitation for correcting reanalyzes or other model-based datasets with observations is the number and quality of the observations themselves, which varies greatly between countries. Specifically, well-maintained and calibrated networks of direct and diffuse solar irradiances, and of turbine rotor plane wind speeds are needed to evaluate and correct reanalyzes for drought studies. If cup or prop-vane anemometers or non-heated sonic anemometers on towers are used in climates where winter temperatures drop below freezing, the data need to be carefully curated for the possibility of icing producing zero wind speeds. Tower shadowing also has smaller, but non-trivial effects. Ideally, one would like to have homogeneous networks of sensors. In addition, to understand the temporal nature of extreme droughts and their return periods, it is desirable that data be collected at the same sites for many years, suggesting a role for government-funded networks, or for rules requiring such measurements from wind and solar plant operators.

An alternative to using meteorological observations to correct reanalysis or other model datasets is to convert the model variables to wind and solar power generation, and then use observed wind and solar power generation statistics provided by grid operators to correct the model-derived values.^30–32 This requires knowledge of the amount of installed wind or solar energy capacity, the amount curtailed, and ideally also what is missing due to maintenance or other issues. Nevertheless, examples for Europe where excellent grid information is available have demonstrated that this method can provide very accurate corrections. In addition to having access to the needed grid-level generation information, a second potential limitation of this method is that it requires renewable generation already to be present in the area analyzed for a time period long enough to provide robust statistics. In some relatively resource-poor regions, that generation may not exist even in developed countries (e.g., wind energy in the southeastern United States) and in countries that currently have little or no renewable generation the method would not be feasible at all.

In a grid system with a significant amount of wind and solar generation, WSDs will be mitigated through a combination of transmission, energy storage, demand response, overbuilding of generation capacity, and especially in present-day systems, using some amount of dispatchable generation (thermal or hydropower). Finding the optimal mix of these mitigation approaches is studied using capacity expansion models, typically with assumed future levels of renewable generation and energy demand. These models require that a given reliability criteria is met, which will typically accommodate the worst-case drought scenarios. For a predominantly renewable energy system, the most difficult periods—requiring the greatest amounts of transmission, storage capacity, storage discharge, and generation capacity—will be during those worst energy balance drought events. If the system is able to meet demand during these events, then it will also be able to meet demand during less-stressed periods with larger generation and/or smaller loads. An improved understanding of how those capacity expansion model solutions relate to drought characteristics could be gained through an analysis of the spatial and temporal characteristics of the drought events. As an example of how droughts can affect capacity expansion model solutions, Grochowicz et al.¹² found that when running a 100% renewable European-scale capacity expansion model, a single week-long drought event can be responsible for up to 77% of the total system build cost.

The complexity and cost of running capacity expansion models have historically limited studies using these models to using short weather data time series (often only one to several years in length) to assess resource adequacy,^33–37 which is too short to capture the range of extreme droughts needed. Recent work has highlighted the importance of including multi-decadal periods.^12,38 In cases with computational limits to running a full multi-decadal simulation, it may be possible to obtain equally accurate results by first identifying drought events, and then running the models over those much shorter drought event periods, although care would need to be taken with this approach as the occurrence of drought events will be a function of the location and type of generation in the portfolio used to initially identify drought events.

Specific mitigation components can include the following:

Building new long-distance transmission allows for the import of power from regions not experiencing droughts. For storage, there are two principal components: storage energy capacity (the size of the storage tank) and storage power capacity (the ability to fill or drain the tank quickly). Sufficient storage power capacity will be needed to meet the most intense load-generation imbalances from droughts on short timescales, while sufficient storage energy capacity will be needed to meet the integrated shortfall of energy of the longest droughts. A wide variety of storage methods and processes have been developed, and new methods are continuously being explored that have the potential to be applicable to WSDs over a wide range of time scales,³⁹ including thermal storage using geothermal reservoirs,⁴⁰ or using molten salt, rock, or other storage media coupled with excess nuclear power generation.⁴¹ 

Demand response programs will be an integral part of electric systems with high shares of weather-driven generation, in part because they are cost effective and can reduce the amount of new transmission and storage built. If designed to incentivize broad participation capable of providing large decreases in load, they can play a significant role in mitigating the impacts of severe droughts. The structure of these programs and decisions by companies or individuals to join them will depend crucially on the statistics of extreme drought events. For some industries, it may make economic sense to agree to dramatically cut back on their electricity use for several days at a time during prolonged drought events, if those events happen infrequently, and if they are offered lower electricity prices during times of abundant generation. In addition, improving demand response program efficacy will require a better understanding of how far, how often, and for how long individuals will cut back on their energy use during energy balance droughts, either based on economic incentives or when asked voluntarily to do so in order to keep the grid stable. For example, the California Independent System Operator (CAISO) has a successful program in which customers are asked to voluntarily cutback on their energy use during times of extreme stress on the grid.⁴² If such requests become more common or of greater duration, will public participation decrease? Multidisciplinary studies involving meteorology, energy, economics, and psychology/sociology may be valuable to understand the limits of demand response programs.

Perhaps the most straightforward way to mitigate WSDs is to build additional wind and solar generation capacity (overbuilding). Many baselines could be used to define overbuilding. One that is simple, easy to understand, and that can be applied to systems with any types of generation is to define overbuilding as the maximum value of the ratio of the instantaneous load to the instantaneous generation. A baseline value of unity would occur if no load or generation variability was present, with the annual mean generation being equal the annual mean load. The maximum of the ratio Load/Generation then gives the multiple of the baseline capacity that is needed to ensure that generation always meets load. Published capacity expansion model studies typically have not quantified their results in terms of the amount of overbuilding needed, instead relying on metrics of the total system cost. We believe that providing results in terms of the capacity overbuilding required would provide a useful additional metric that would facilitate comparisons between different studies and different solutions.

The most common approach currently used for mitigating WSDs is through dispatchable generation, which can be broadly classified as either thermal generators (often fossil fuel based) or hydropower. Geothermal and nuclear power are two low-carbon thermal generators typically included in capacity expansion model studies as baseload-generation, providing power at constant levels. However, geothermal plants could potentially be operated as a dispatchable power generator via intermittent operation,^43,44 perhaps only used seasonally or during extreme energy balance droughts. Intermittent operation may be desirable for geothermal electricity generation, especially from retro-fitted oil and gas wells, from which production can rapidly decline over time due to a decrease in temperature gradient and pressure as heat is extracted.^45–47 A potential benefit of operating these wells in a dispatchable mode would be to allow ground temperatures to recharge when geothermal is not needed, so that they could generate maximum power when needed most. In addition, it may be possible to use geothermal as an energy storage medium, transferring heat into aquifers during periods of excess electricity generation.⁴³ Also, operating geothermal generation on an intermittent basis could potentially enhance its economic value.⁴⁴ 

Although most often run as baseload generators, nuclear power plants can also be run in a dispatchable mode. For example, nuclear power plants in France typically have the ability to perform two load following operations per day at 5% of power per minute, and can go down to as low as 20% power. In addition, some plants are shut down in summer when energy demand is low.⁴⁸ However, a compounding uncertainty is that nuclear plant generation availability can also be dependent on weather. That is, during heat waves, limits on access to sufficiently cold water for cooling can lead to curtailment of nuclear generation.

Finally, thermal powerplants run on biofuel or synfuel have been proposed as a complement to renewable generation for a low- or zero-carbon electrical grid.⁴⁹ In either case, fuel would need to be stored to be available during periods of WSDs, and accurate characterization of the climatology of WSDs would be crucial for determining the required fuel storage capacity for such systems.

Hydropower is the other main type of dispatchable generator. However, unlike thermal generators, hydropower often operates under a set of constraints that can limit its ability to generate power when needed. These constraints include restrictions on minimum streamflow for irrigation, flood control, ecological purposes, and recreation. In addition, hydropower generation can be restricted during extended periods of hydrologic drought. Hydropower drought is related to hydrologic drought (a deficit in streamflow or water storage), although the relationship can be complex and nonlinear due to human management of natural water systems. Hydropower drought is not often studied alongside wind and solar drought due to the longer time scales involved and the relative complexity of hydrologic modeling. A few studies have examined hydropower droughts and their co-variability with wind and solar droughts,^5,50,51 but more study is needed. The complementarity between hydro, wind, and solar is highly regional and also depends on the specific design of dams or diversion facilities relative to the availability of water. Smaller facilities are more influenced by weather variability but there is still a lag time between the precipitation, streamflow, and hydropower output. Larger hydropower facilities may have the ability to withstand even multi-year hydrologic droughts with limited impact on hydropower. In colder regions, snowpack will also influence the time scales of hydropower droughts. At all timescales, hydropower droughts and their linkage to wind and solar droughts should be considered as part of studies on long duration energy storage.

What are the implications of wind and solar energy droughts when determining the necessary energy storage and generation capacity needed to reliably operate an electric grid with high shares of renewables? To address this question, and to highlight the relation between WSDs, transmission, and overbuilding, we performed the following analysis of an idealized grid scenario. The location (Fig. 1) and capacity of existing (as of August 2024) wind and solar projects was identified using the U.S. Geological Survey wind turbine and solar farm databases (available at https://eerscmap.usgs.gov/uswtdb/ and https://eerscmap.usgs.gov/uspvdb/, respectively). Wind and solar generation were calculated using bias-corrected ERA5 wind and solar generation values²⁷ interpolated to the plant locations for the 63-year period between 1959 and 2021. Load was defined as a linear function of temperature depression below 14 °C and temperature elevation above 20 °C (the results weakly depend on the exact choice of the load-temperature dependency) and calculated with equal weighting at each location where there was a wind or solar plant using the raw ERA5 values for 2-m temperature. This represents a hypothetical future scenario where heat pumps are used for building heating, and future wind and solar plants are built with the same geographical distribution as they currently have. Once the mean load (averaged across all 63 years) and the wind and solar generation were known for each averaging radius relative to the central location denoted by the red dot in Fig. 1, the wind and solar generation aggregated over that region was multiplied by a factor such that the mean of each generation type would equal half of the mean load averaged over the same region. Daily values of wind and solar generation and load were then analyzed for different aggregations in time and space. The analysis shown here is for the month of January, when loads on average for the CONUS will be the largest, and overall generation is relatively low (because of the lower solar contribution).

Figures 2(a) and 2(b) show how minimum values of daily averages of wind and solar generation for the month of January, normalized by their respective radius-dependent 63-year annual mean generation values, scale with temporal and spatial averaging. Spatial averaging is equivalent to having strong transmission over that area, while temporal averaging is equivalent to having storage that can smooth out variations over time scales shorter than the averaging length. The minimum values of wind and solar generation can be extreme, close to zero, for short averaging periods and small averaging areas, and increase significantly for averages taken over several weeks and a radius of nearly a thousand kilometers. Starting from the smallest radius and averaging time, solar and wind minimum values initially increase at similar rates, both temporally and spatially, but solar then plateaus to values around 0.5–0.6 for the longest averaging times and largest radii, while wind keeps increasing to a value above 0.7. For averaging lengths up to about a week, for both wind and solar, the change due to ∼4 days of temporal averaging is equivalent to spatial averaging with a radius of ∼600–800 km.

Normalized load (relative to the annual mean load) is shown in Fig. 2(c) and normalized net-load (load minus generation, also relative to the annual mean load) in Fig. 2(d). For short averaging periods and small averaging areas, the maximum load is approximately twice as large as the mean load, while over larger averaging areas and time intervals the ratio trends lower. For small temporal and spatial averaging, the maximum net-load is about a factor of 1.6 times greater than the mean load, while for the largest time and spatial averages, maximum net-load remains strongly positive, with a value of ∼0.8. (A value of unity corresponds to the net-load being equal to the 63-year mean load, while a value of zero corresponds to the generation and load being in balance.) This demonstrates that even with large balancing areas with strong long-distance transmission, long-duration storage, and/or building generation capacity above that needed to match the mean load is a necessity in this illustrative example. This is of course true of a fossil fuel system as well due to the temporal variability of the load.

Insights into the impacts of droughts on overbuilding of generation capacity as well as storage can be drawn by considering the ratio of load to generation, as shown in Fig. 3. For a renewable energy system with no storage or demand response, the maximum value of Load/Generation gives the amount of overbuilding needed in order for generation to always meet demand. For the smallest temporal and spatial averaging, the ratio indicates that generation would have to be overbuilt by a factor of about 8 times greater than the mean load. For spatial averaging with a radius of 800 km, the overbuilding is reduced to a factor of 4. The ratio also decreases rapidly with temporal averaging, such that by 1 week at a small radius, or 3–4 days at a large radius, the overbuilding required drops to a value of 2.

As a consequence, relatively modest amounts of storage capacity (∼7 days of mean load for the smallest radius, 2–3 days for a radius of 400 km) would be sufficient to reduce the overbuilding needed to roughly a factor of 2. However, the maximum rate at which that storage would need to be discharged is given by the net-load [Fig. 2(d)] for the shortest temporal average of one day. Those values are large: 1.5 times mean load for a radius of 200 km, and 1.1 times mean load for a radius of 900 km. Additional mitigation measures, including demand response and dispatchable generation would be needed to reduce those large storage discharge rates.

Periods of time when wind and solar generation are well above their long-term average can present either opportunities for charging storage or challenges for grid operations due to transmission congestion or curtailment of generation. In addition, in many grid balancing areas with large amounts of wind and solar generation, energy prices often become negative during flood events.⁵² It can be important therefore to understand the statistics of these surplus events. Kittel and Schill⁹ discuss some of the challenges of analyzing “positive residual load events” in ways that best support system design. Following Wilczak et al.,¹ we refer to these high production periods as “floods,” where “flood” is used in the sense of a state of abundant flow or volume, which can have either positive or negative connotations.

In a future energy system that may be highly renewables based, will all of the energy produced when wind and solar are both generating near their highest levels be utilized, or will some of it be curtailed? For example, the largest 1-day flood events for different regions across the United States can exceed 200%–400% of the long-term mean wind or solar generation.¹ Since making use of that energy requires building infrastructure such as intermittent manufacturing or storage charging capacity that would be used very rarely, it seems more likely that some if not much of the generation during the periods with exceptionally high generation will be curtailed. However, more frequently occurring periods of weaker positive anomalies may be able to be used effectively for either charging of storage or by increased intermittent industrial use.

Actions taken to mitigate low energy storage episodes, such as demand response, will depend on the current state of that energy storage and forecasts of how low storage will get and for how long, based on short-term to sub-seasonal to seasonal weather forecasts. Identifying weather patterns associated with energy balance droughts, as discussed above, has the potential to help in forecasting those droughts on time scales of days to weeks.⁵³ Identifying sea surface temperature (SST) patterns or climate modes of variability associated with energy balance droughts, discussed below, has the potential to help in forecasting those droughts on longer time scales. Predicting extended droughts will also be important for planning of maintenance for all types of generation including large baseload plants.

Currently, little is known of the predictability of energy droughts across the range of relevant timescales. Of great utility would be developing a deeper understanding of the limits to predictability of energy droughts at all timescales in current forecast systems—for example, determining whether wind droughts are more (or less) predictable than solar droughts at varying lead times, as found by Li et al.⁵⁴ in a case study of a WSD event in the Netherlands, and whether compound wind, solar, and load extreme events have different predictability characteristics than each on their own.

Modes of climate variability are naturally occurring quasiperiodic states of the climate system, many of which are known to have profound impacts on regional- to global-scale weather patterns. Several of the most dominant of these modes are, from the shortest to longest periodicities,

• the Madden-Julian Oscillation (MJO) which is the largest element of the intra-seasonal (30–90 day) variability in the tropical atmosphere, and is characterized by an eastward progression of large regions of enhanced or suppressed cloudiness and tropical rainfall, observed mainly over and between the Indian and Pacific Ocean;

• the North Atlantic Oscillation (NAO), an east–west variation in north Atlantic sea-level pressure with an irregular, year to multi-year periodicity;

• the El Niño Southern Oscillation (ENSO), characterized by an approximately 2–4 year cyclical variation in tropical Pacific sea surface temperatures and zonal pressure gradients;

• the Pacific North American (PNA) teleconnection, a variation in eastern Pacific and North American surface pressure gradients, operating over a range of time scales from annual to decadal;

• the Pacific Decadal Oscillation (PDO), determined by there being warmer or cooler than average SSTs in the Pacific north of 20°N, with each phase lasting 20–30 years;

• and the Atlantic Multi-decadal Oscillation (AMO), consisting of long-duration changes in SSTs in the North Atlantic Ocean, with cool and warm phases that may last up to 20–40 years.

Tropical solar insolation will obviously vary with the MJO; surface wind speeds in Europe have also been shown to depend on the phase of the MJO;⁵⁵ wind power in Europe has been shown to be correlated with the phase of the NAO;^51,56 ENSO has been shown to have significant impacts on wind energy over Africa⁵⁷ and the United States;⁵⁸ multiple modes of climate variability (NAO, ENSO, AO, and PDO) have been linked to variations in hub-height wind speeds across North America, dependent on season and location;^59,60 surface wind speeds globally and on continental scales have been shown to vary on decadal time scales, with those variations accurately reconstructed using combinations of several climate modes of variability;⁶¹ and the PDO has been shown to be correlated with solar irradiance in the United States.⁶² These quasiperiodic, low-frequency modes of variability, when superimposed on random, short-term weather-scale to sub-seasonal variations, may then determine the amplitude and timing of the most intense drought events. Low-amplitude but long-duration droughts have the potential to affect the optimal amount of energy capacity of storage systems, as determined by capacity expansion models, while the superposition of these modes with short-term weather fluctuations may determine the optimal power capacity of energy storage systems. A recent report by the Royal Society³⁹ explored the potential for multi-year renewable energy droughts over the United Kingdom (e.g., a challenging winter followed by a low-renewable summer that does not allow time for a full recovery of large-scale storage facilities). These multi-year events have the potential to significantly increase large-scale storage requirements. A better understanding of the impacts of each of these climate modes of variability on renewable energy production is needed, especially their regional impacts, as well as the impact of the various modes acting in concert.

How will the statistics of WSDs, and load, change under different climate change scenarios? Can impactful WSDs shift to periods of larger loads (more extreme temperatures) in a warmer climate? These are challenging questions as the range of possible responses of wind and solar power to climate change vary depending on factors such as jet stream positions or global precipitation patterns, and these may vary depending on the rate and magnitude of the warming (IPCC, 2021).

Examining changes in global severity of wind droughts from 1979 to 2022, a recent study by Antonini et al.³⁸ does not find increases in wind drought severity through time. Similarly, van der Weil et al.¹⁴ examined both energy supply and energy balance droughts in Europe using two climate models, finding that climate variability has larger impacts on wind and solar generation than climate change, and that in a warmer climate scenario there is no change in the occurrence of extreme low energy production events. Pryor et al.⁶³ found that to 2050 there was no change in wind drought frequency, while Coburn and Pryor⁶⁴ found that by 2099 an increase occurred in several climate models, with that increase due to a slight weakening of the overall mean wind speed. Kapica et al.⁶⁵ evaluated changes in the number of WSD days across Europe in a suite of seven regional climate models under two future climate scenarios, where droughts are defined as days when the renewable generation is less than a threshold percentile. They find that the duration of solar droughts increases in northern Europe but decreases in southern Europe, while for wind droughts the results are less spatially consistent. Moreover, there are many regions where there is a lack of agreement between the models.

When considering future climate projections across the 6th Coupled Model Intercomparison Project (CMIP6) archive, with 2 °C of warming the area most strongly impacted by reductions in near-surface wind speeds is central Europe (IPCC, 2022). However, some regions such as the central United States, East Africa, central America, and Brazil see increases in near-surface winds. All of these results are highly uncertain, with most of the globe showing low agreement between models. Precipitation changes (which will influence solar and hydropower generation) are also highly uncertain, with very low confidence over most of the global land area in the sign of any future changes across the CMIP6 archive (IPCC, 2022). This leaves the potential impacts of climate change on renewable energy droughts highly ambiguous, and more study is needed to assess the impacts of climate change on WSDs across the globe.

Wind and solar energy generation is growing rapidly not only in developed countries but also in economically developing countries, where population and energy demand likewise are often increasing rapidly. Building efficient and reliable electric energy systems dependent on renewable energy would benefit greatly from a thorough understanding of wind and solar energy droughts in all locations. Especially in developing countries, this will require building new observational networks to validate and correct model-based datasets. Without these investments, the range of science issues discussed in this Perspective article will be much harder to accomplish in less industrialized countries.

Energy balance droughts can always be mitigated if through informed planning sufficient generation, transmission, and storage of energy is built. The question is whether the cost of such a system is acceptable to society. To date, capacity expansion models have shown mixed results on the net cost for highly renewables-based systems, some showing lower costs and some higher than reference scenarios representative of the present-day grid⁶⁶ with the caveat that those capacity expansion models have typically been run over relatively short (<7 years) time spans that may not adequately sample the full range of drought magnitudes. Qiu et al.⁶⁶ attributed those differences to assumptions made concerning demand response and flexible loads, whether the build out of the renewables system is gradual, and whether the social costs of carbon emission and other pollutants are accounted for.

As electric grids become more renewables-based simply because wind and solar are the cheapest sources of new generation, they will become increasingly vulnerable to renewable energy droughts unless new long-distance transmission and multi-day storage is built or sufficient dispatchable generation sources are maintained. Unlike hydrologic droughts, which can produce severe societal impacts even with good planning, solutions to wind and solar energy droughts in highly renewables-based energy systems exist, such that their occurrence would have a manageable impact on the average energy user. In particular, mitigation of these droughts can be achieved through integrated planning of combinations of new energy transmission, storage, dispatchable resources, and demand response programs. However, achieving this goal will require improving our understanding of the spatial and temporal characteristics of these droughts. Specific solutions will depend not only on the relative costs of transmission, storage, and demand response but also on the intensity, duration, spatial extent, and frequency of occurrence of energy droughts.

Several areas of research have been highlighted that we believe would greatly improve our understanding of how significant wind and solar droughts can be and of their impacts on the energy system. We note that some of these areas are already actively being researched in regions with considerable renewable energy deployment like Europe, the United States, China, and Australia. However, there are many regions where research on these topics has not yet started. This is particularly true in economically developing countries, where access to high-quality meteorological and energy data are a greater challenge. In all regions, the characterization of WSDs could contribute to the successful planning and operation of new renewable generation, as well as understanding the challenges to existing generation. In summary, these key areas of research include the following:

1. Linking capacity expansion model solutions to drought characteristics.

2. Classifying meteorological regimes under which energy droughts occur.

3. Evaluating the accuracy of solar droughts determined from the NSRDB and SARAH datasets.

4. Characterizing wind, solar, hydropower, and load compound events through their intensity, duration, and frequency of occurrence statistics for different regions of the world.

5. Determining drought predictability at varying lead times, for wind or solar, and for wind, solar, and load compound events.

6. Determining the potential for snow and ice on PV panels to create solar droughts.

7. Determining how much curtailment will occur during renewable resource flood events, and how will this vary by region and the wind/solar/storage mix.

8. Characterizing WSD statistics for their impact on demand response programs.

9. Understanding the impacts of climate modes of variability on WSDs, including for multiple modes acting in concert.

10. Determining how the statistics of WSDs, and load, will change under different climate change scenarios.

11. Applying drought analyses to countries outside of Europe and North America, especially in developing countries.

Finally, there is a clear need for better datasets upon which energy drought analyses are based. This includes more accurate reanalysis models run at higher spatial resolution, and the creation or expansion of networks observing the direct and diffuse solar irradiances, and turbine rotor height wind speeds.

The authors thank Delavane Diaz, Dan Livengood, and Mike Hobbins for helpful reviews of the manuscript. The views expressed in this paper are those of the authors and do not necessarily reflect those of NOAA or EPRI or its members.

J.M.W. was funded by the NOAA Physical Sciences Laboratory and the NOAA Atmospheric Science for Renewable Energy program. H.B. was funded by a Newcastle University Academic Track Fellowship. C.B. was funded through the Grid Operations, Decarbonization, Environmental and Energy Equity Platform (GODEEEP) Investment, under the Laboratory Directed Research and Development (LDRD) Program at the Pacific Northwest National Laboratory (PNNL).

The authors have no conflicts to disclose.

James M. Wilczak: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Supervision (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Daniel B. Kirk-Davidoff: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Resources (equal); Software (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal). Hannah Bloomfield: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Resources (equal); Validation (equal); Writing – original draft (equal); Writing – review & editing (equal). Cameron Bracken: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Resources (equal); Validation (equal); Writing – original draft (equal); Writing – review & editing (equal). Justin Sharp: Conceptualization (equal); Formal analysis (equal); Funding acquisition (equal); Investigation (equal); Methodology (equal); Resources (equal); Validation (equal); Visualization (equal); Writing – original draft (equal); Writing – review & editing (equal).

The data that support the findings of this study are openly available in U.S. Geological Survey at https://eerscmap.usgs.gov/uswtdb/ (Ref. 67) and https://eerscmap.usgs.gov/uspvdb/ (Ref. 68).