Transforming fossil-fuel-based energy systems to rely on renewables is essential to reduce greenhouse gas emissions and mitigate climate change^1,2,3. Wind and solar energy have become mature and economically competitive^4,5,6, but they are variable and intermittent^7,8,9. This variability, intermittency, and resource uncertainty pose a challenge to investors, system operators, and policymakers as they seek to employ an appropriate mix of technologies and strategies to meet the economic, reliability, and environmental goals of the electric power sector^10,11,12. Extreme or prolonged reductions in renewable power generation (hereafter, renewable energy droughts) create electricity shortage risks and threaten the reliability of power system operations^13,14. In some cases, wind and solar energy droughts coincide across large regions, causing severe impacts on energy systems^15,16. If these renewable energy droughts are not carefully characterized, they may threaten efforts to accelerate renewable energy deployment by sapping investor, public, or policy support for these technologies.

Although numerous studies have examined the impact of renewable energy droughts on power systems^17,18,19, there is no unified definition for renewable energy droughts. Raynaud et al.²⁰ defined it as continuous low power production or mismatches with demand, while Rinaldi et al.²¹ defined it as a solar or wind capacity factor less than 50% of the daily mean for that day of the year. Kapica et al.²² built upon the definition as a capacity lower than a threshold and proposed measuring the duration of the drought as the number of time periods it occurs, represented by a sum of binary variables. Allen and Otero²³ proposed standardized indices based on quantiles rather than a percentage of the mean, defining droughts by percentiles and their magnitude by the sum of the drought index over a certain period. Antonini et al.²⁴ proposed an energy deficit metric that integrates the depth and duration of low wind power, using power density, seasonal variability, and weather variability to identify reliable sites for wind power generation. Building on these, we characterize energy droughts by calculating deviations from expected values, crucial for grid planning, and using quantile-based thresholds and binary markers to analyze spatial and temporal patterns across different renewable resources.

Current knowledge about wind and solar energy droughts is limited, including a lack of understanding of the extent to which spatial and temporal coincidence exacerbates their impacts. Research has noted an increased frequency and severity of extreme climate and weather episodes^25,26,27. These studies have paid attention to coincident extreme events, such as wildfires, droughts, and heatwaves^28,29,30 and their compound or cascading impacts; some studies have also focused on the spatial distribution of weather extremes^31,32,33,34 and their spatial co-occurrence^35,36,37,38. These extreme weather events could lead to low renewable energy generation^39,40; some—for example, continuous windless or overcast weather—could lead to long spells of low wind and solar energy generation⁴¹. They may affect large geographic areas at the same time and affect both wind and solar, which could extend their detrimental impacts beyond limited disruptions to energy system reliability and affect social and economic welfare.

The scientific literature has characterized the resource quality using the resource potential and variability of wind and solar energy in different countries. Studies have estimated the wind energy resource potential worldwide⁴² and in specific countries such as China⁴³; some have also quantified its spatiotemporal variability^44,45. Other work has assessed global solar energy potential and its spatiotemporal variability⁴⁶. A few studies have quantitatively compared the spatiotemporal variability of wind and solar energy and found complementary patterns^47,48. Overall, the literature has provided essential insights into wind and solar potential and variability in different locations. Still, there is a fundamental literature gap in understanding the resource drought of different sites. There are differences between the impacts of resource variability, which can be managed with the installation of short-duration battery energy storage, and the impacts of resource droughts, which require backup generation and long-duration energy storage resources to prevent blackouts. Given this difference, there is a lack of a metric to describe the resource quality in terms of its propensity to energy droughts. We term this attribute of proneness to droughts as “extremeness”.

We seek to fill this gap by offering a detailed characterization of wind and solar energy droughts and their spatial and temporal coincidence. We propose and estimate novel metrics to assess a site’s renewable energy quality, accounting for its propensity to experience droughts and the likelihood that these occur over large geographic areas and across resources, with coincident wind and solar droughts. We make three contributions. One: we propose a statistical method to identify and characterize renewable energy droughts using long time-series data at high spatiotemporal resolution. We account for the intensity and duration of extremes and identify three categories of renewable energy droughts that pose different risks to the power system. With this method, we depict the spatial pattern and temporal distribution of renewable energy droughts. Two: we propose a method to examine the prevalence of spatial and temporal coincidence of these droughts. We determine in which regions these droughts are spatially correlated, jeopardizing large geographic areas. We also identify the regions and seasons where and when wind and solar energy extremes are likely to occur simultaneously. Three: we propose new metrics to characterize renewable energy quality along three dimensions (availability, variability, and extremeness). The availability measures how much of the resource can be utilized at the site, as indicated by its average capacity factor. The variability measures how stable the resource is as indicated by the standard deviation of the capacity factor. This metric can guide the development of short-duration energy storage. The extremeness is proposed as a new dimension of resource quality to measure how likely the site is to experience energy droughts. This metric can guide the development of long-duration energy storage and other backup resources. By using this triad of metrics, we improve existing understanding of renewable energy quality and identify “ideal” sites for renewable energy development.

The proposed metrics are applied to identify and characterize renewable energy droughts in China, the country with the world’s largest electric power sector. As nations around the globe explore pathways to transition to deeply decarbonized electric power systems, the method presented will guide analyses that characterize renewable energy droughts and assess renewable energy quality to further develop these electricity sources. The results could prove essential to inform capacity expansion and operation strategies for a reliable decarbonized energy system insulated from the perils of wind and solar droughts.

We propose using a triad of indicators to assess renewable energy quality: availability, measured with the annual capacity factor; variability, measured with standard deviation; and extremeness, measured with the RDF index, an index we propose that measures the “droughtiness” of the resource. This droughtiness, which we term “extremeness”, measures the severity and frequency of episodes when wind power, solar power, or both are extremely low relative to their expected value.

While availability and variability are measured with commonly used metrics, assessing extremeness requires us to define a drought. We define a drought as an episode when the resource’s performance significantly deviates from its statistical expectation (more in the Methods section). Because renewable energy droughts differ in their severity, we define three categories of drought (blue, orange, and red), each one denoting a higher level of severity. We analyze over 19 years (2000–2018) of hourly data on wind and solar resources in each grid cell of 0.5-latitude-degree and 0.625-longitude-degree to identify past periods of droughts and determine their coincidence across resource (i.e., wind or solar), time, and space.

In what follows, we show the spatial and temporal distributions of wind and solar energy droughts; their spatial correlation and temporal coincidence; and renewable energy quality considering resource availability, variability, and extremeness.

Occurrence of renewable energy droughts

Figure 1 shows the temporal distribution of renewable energy droughts across the 24 h of a day and the 12 months of a year using the Relative Drought Frequency (RDF) index (see Methods). Three major findings are discussed.

The vertical axis shows the 12 months of the year, while the horizontal refers to the hours in a day. A–C Relative Drought Frequency (RDF) index of wind energy droughts which can be compared with (D), the capacity factor (CF) of wind energy; E–G RDF index of solar energy droughts which can be compared with (H), the CF of solar energy.

Full size image

First, high capacity factor (CF) periods are coincident with periods experiencing a high likelihood of renewable energy droughts. By comparing the temporal distribution of renewable energy droughts (Fig. 1A–C, E–G) with the resource availability of wind and solar (Fig. 1D, H), we can see that the seasonal and daily patterns of renewable energy droughts are similar to the distribution patterns of resource availability. For instance, wind energy is plentiful during the spring and winter, and wind droughts are also frequent in these periods; also, solar energy resources are abundant at noon in spring and autumn, and the solar droughts are concentrated during this period. During the high CF period, it is important to be aware of likely reductions in generation and unexpected droughts that can occur, creating electricity shortfall risks.

However, not all renewable energy droughts occur at periods with high resource availability. For example, wind energy is richest in April, but this month is not the highest occurrence period of wind droughts; solar energy is rich in summer, but this season has the least solar droughts. In comparison, solar droughts have the highest occurrence in November, but this month has relatively low solar energy resource; also, wind droughts have the highest occurrence in May, but wind is not most abundant this month. During these periods with high drought occurrence and low resource availability, power system operators could address resource scarcity issues and extreme low-energy events by seeking alternatives, such as firm and low-carbon generating resources, short-duration and long-duration energy storage systems, and demand response.

Second, the seasonal patterns of wind and solar energy droughts are different, but spring is a time when both wind and solar droughts occur. Wind droughts mainly occur in spring (March, April, and May); a smaller number of droughts occur in winter (December, January, February). Red droughts, which are most severe with high intensities and long durations, are more frequent in spring compared with the other two categories of wind droughts. May has the highest occurrence of all three categories of wind droughts; autumn has almost no red droughts. In comparison, solar droughts are relatively evenly distributed across all seasons. Solar droughts are more frequent in spring and autumn; summer has a relatively low occurrence of solar droughts. More severe droughts are more common in late autumn (November).

Third, wind and solar droughts occur at different hours of the day but sometimes coincide at noon and in afternoons. Wind droughts are distributed throughout the day. Although they are more likely in the afternoon (12:00–16:59), they do occur at night, too (22:00–02:59). Red wind droughts are more prevalent around noon than the other two categories. Solar droughts mainly occur at mid-day (11:00–13:59); and many droughts also occur in the afternoon (14:00-15:59), coinciding with wind droughts. The more severe solar droughts become, the more likely they are to occur at noon.

Figure 2 shows the spatial distribution of renewable energy droughts. Wind and solar droughts tend to be concentrated in areas which, generally, have higher resource availability. The more severe droughts (turning from blue to orange and then to red) are more prevalent in areas such as Inner Mongolia, Xinjiang, and Qinghai. In comparison, areas with lower resource availability, such as Guangdong, Jiangsu, and Zhejiang, have fewer renewable energy droughts; these provinces are spared the most severe (red) droughts.

A Relative Drought Frequency (RDF) index and capacity factor for each wind energy eligible site; B RDF index and capacity factor for each solar energy eligible site.

Full size image

While wind droughts are concentrated in the north, solar energy droughts are more distributed across regions. For example, in Inner Mongolia and Xinjiang many sites have no red wind energy droughts, but in these provinces almost all sites have red solar energy droughts. Also, the maximum RDF value for wind energy is higher than that for solar energy, which once again indicates that wind energy droughts are more prevalent in some sites.

Spatial and temporal coincidence of renewable energy droughts

Figure 3 shows the coincident wind-solar energy droughts. The three categories of coincident droughts have similar temporal distributions (Fig. 3A–C): spring has the most coincident droughts, followed by winter. Summer has the least coincident droughts. More severe droughts are more likely to occur in spring and winter, especially in January, March, May, and December. Coincident droughts during December and January are especially concerning because these two months could also exhibit high electricity demand once residential space heating is further electrified.

A–C Temporal distribution of the Relative Drought Frequency (RDF) index for coincident droughts across the 24 h of a day and the 12 months of a year. D RDF index of coincident droughts in each province; the order of provinces is arranged by the RDF value of blue droughts.

Full size image

Despite this, there are differences in the temporal distribution of the three categories of coincident droughts. Blue droughts are more frequent at mid-day (11:00–13:59) in May. In comparison, red droughts are more prevalent at mid-day (10:00–14:59) in March. No red droughts occur during the summer, indicating that the most severe coincident droughts are unlikely to occur in the summer. This is good news for the Chinese power system where space cooling demand in the summer causes a peak in electricity consumption. In addition, the temporal distribution of coincident droughts is different from the distribution of either wind or solar energy droughts.

In terms of spatial distribution, coincident wind-solar droughts present distinguishably different occurrence patterns compared with either wind or solar droughts (Fig. 3D). The resource-rich areas, such as Xinjiang, Inner Mongolia West, Qinghai, and Gansu, have numerous wind or solar droughts but a very low prevalence of coincident droughts. In comparison, southern provinces such as Fujian and Sichuan have a high prevalence of coincident droughts. The results indicate that the development of wind energy in these resource-rich areas should go hand in hand with the exploitation of solar resources to take advantage of the scarcity of coincident wind-solar droughts, while the simultaneous development of wind and solar energy in Fujian and Sichuan might not mitigate the risks of electricity shortages.

Figure 4 shows the spatial dependence of renewable energy droughts. Clustering analysis on provinces and co-occurrence network analysis on sites reveal the strong spatial dependences of renewable energy droughts. Clustering reveals that there is a positive correlation in the number and severity of renewable energy droughts across geographically adjacent provinces (Fig. 4A, B). Most groups of provinces with strong spatial correlation of droughts are connected to the same regional power grid, others are connected to adjacent power grids. The co-occurrence network analysis also reveals that sites in the same province, same power grid, or adjacent power grids are more likely to have concurrent renewable energy droughts (Fig. 4C, D). For example, there is high correlation in the wind and solar droughts occurring in the provinces of Heilongjiang, Inner Mongolia East, Jilin, and Liaoning, which are all connected to the Northeast China Grid. Similarly, there is high correlation in the solar droughts of Tianjin, Beijing, Hebei, Henan, Jiangsu, and Shandong, which are all connected to adjacent power grids (North China Grid, Central China Grid, and East China Grid) that might call on one another in a future, net-zero China with a large interconnected transmission system.

A, B Spatial correlation matrix and clusters of wind and solar blue droughts. We use the Pearson correlation coefficient to measure the spatial correlation of wind and solar energy droughts among provinces. Each coefficient reflects the strength and direction of the linear relationship between the occurrences of droughts in different provinces. A positive value indicates that the two provinces are positively correlated in terms of the prevalence of renewable energy droughts, suggesting that they tend to experience such events concurrently. Conversely, a negative value implies a negative correlation. C, D Co-occurrence networks of wind and solar blue droughts. Each of the nodes in the graphs represents sites with droughts. The links connecting two nodes indicate that their Jaccard similarity coefficient of renewable energy droughts exceeds 0.5, which means that droughts in those two sites co-occurred during at least half of all hours.

Full size image

More importantly, for most provinces, the occurrence of renewable energy droughts is positively correlated with their occurrence in other provinces; only a small number of provinces have negative correlations. Positive spatial correlation of solar droughts is more prevalent than wind. There is not a single pair of provinces with a negative correlation of solar droughts. This result indicates that renewable energy droughts, especially solar droughts, are spatially coincident across vast regions, so that the solar resources of one province can almost never be relied upon to withstand a drought in another. For a few provinces, such as Fujian, Guangdong, Guangxi, and Hainan, the correlation of wind droughts is negative. Since wind energy droughts in these southern provinces and other provinces do not occur simultaneously, the resources can be complementary in time and wind capacity expansion planning can account for the benefits of long-distance transmission lines in reducing electricity shortfall risks.

Figure 5 shows the spatial dependence of coincident wind-solar energy droughts. There is a strong correlation of coincident wind-solar energy droughts among geographically adjacent provinces. Unfortunately, the correlation is stronger in the northern provinces which have more renewable energy droughts. Compared with other provinces, Beijing, Hebei, and Tianjin consistently show the strongest correlations of the three categories of coincident droughts. Heilongjiang, Inner Mongolia East, Jilin, and Liaoning; or Ningxia, Shaanxi, and Shanxi also consistently show strong correlation of the three categories of coincident droughts. The results mean that a coincident wind-solar energy shortage may occur simultaneously in multiple provinces in the same power grid or even across several adjacent power grids, threatening operational reliability. The frequency of large-scale coincident shortages is higher in the northern power grids, such as in the Northeast China Grid, North China Grid, and Northwest China Grid, highlighting potential difficulties in maintaining system reliability within these grids.

It shows the spatial correlation matrix and clusters of coincident blue droughts. We use the Pearson correlation coefficient to measure the spatial correlation of coincident wind-solar energy droughts among provinces. Each coefficient reflects the strength and direction of the linear relationship between the occurrences of coincident droughts in different provinces. A positive value indicates that the two provinces are positively correlated in terms of the prevalence of coincident droughts. Conversely, a negative value implies a negative correlation.

Full size image

Metrics of renewable energy quality

We propose the simultaneous use of three indicators to comprehensively assess renewable energy quality: availability, measured with the annual CF; variability, measured with standard deviation; and extremeness, measured with the RDF index. High renewable energy quality—of a geographical site— requires high availability, low variability, and low extremeness. Figure 6 presents the wind and solar energy quality observed in China.

A wind energy eligible sites; B solar energy eligible sites. Each circle represents a site, with its size indicating the relative magnitude of resource availability measured as capacity factor. The circle’s color indicates the resource availability relative to the average value. Purple indicates a site’s capacity factor is higher than the average across all sites, while gray indicates it is lower.

Full size image

First, the quantification of energy quality demonstrates that site selection for renewable energy development presents a trilemma. The data show that no sites exhibit the three desirable qualities concomitantly: high availability, low variability, and low extremeness of wind or solar energy resources. As the resource availability increases (i.e., as it changes from gray to purple dots in Fig. 6), variability and extremeness also increase (i.e., they go from the bottom left to the top right corner in each panel in Fig. 6). This indicates that the availability, variability, and extremeness of renewable energy present themselves in undesirable directions in the region observed. Sites with higher resource availability (the purple dots in Fig. 6) generally have higher variability and more energy droughts; while areas with lower variability and fewer droughts generally have lower resource availability (the gray dots in Fig. 6). For example, many sites in Inner Mongolia and Xinjiang have higher resources availability and also higher variability and more energy droughts; while sites in Guangdong, Jiangsu, and Zhejiang have lower availability, lower variability, and fewer energy droughts.

Second, a small number of sites have higher variability but lower extremeness than average (dots in the top left corner of each panel in Fig. 6); most of these sites have higher wind or solar resource availability than average (i.e., most are purple dots). In contrast, a smaller number of sites have low variability but high extremeness (dots in the bottom right corner of each panel in Fig. 6); most of these sites have lower resource availability of wind or solar energy than average (i.e., most are gray dots). By comparison, sites in the top left corner outperform those in the bottom right corner in terms of renewable energy quality, because the former have both higher resource availability and lower extremeness, while the latter have higher extremeness and lower availability.

Third, only a minimal number of sites rank slightly above average in the three renewable energy quality dimensions, i.e., have high availability, low variability, and low extremeness (i.e., are purple dots in the bottom left corner of each panel in Fig. 6). Only 22 out of 773 suitable sites (2.8%) have higher wind resource availability, lower variability, and fewer energy droughts than the average; half of these are located in Inner Mongolia. In contrast, only 1% of sites (16 out of 1663) have solar energy with higher availability, lower variability, and lower extremeness than levels; most of these sites are in Xinjiang.

Fourth, wind and solar differ in terms of quality. Wind energy generally has higher resource availability than solar, but it also has higher variability and extremeness. However, there are more sites with an overall acceptable wind energy quality than solar. Most of the sites with high solar availability also have high variability, while more sites have high wind resource availability combined with low variability. This does not even consider the diurnal nature of solar energy, which is also an important disadvantage of this resource relative to wind.

As the power sector transitions towards decarbonization, the importance of the three key dimensions—resource availability, variability, and extremeness—shifts (Fig. 7). At low penetration levels of solar and wind resources, resource availability is the most critical metric. With low shares of electricity being generated by these sources, the system retains sufficient flexibility to manage their inherent fluctuations while satisfying the electrical demand. In this context, policymakers and project developers prioritize increasing renewable energy production and reducing the Levelized Cost of Electricity (LCOE), which is largely dependent on the capacity factor or resource availability, with limited concern for system flexibility. As the penetration of variable renewable energy increases, system planners shift their focus to resource variability. They could strategize how to enhance system flexibility through the deployment of dispatchable generation, short-duration energy storage, and power transmission infrastructure. At deeply decarbonized systems, the risk of energy droughts becomes a critical reliability concern. Consequently, planners could concentrate on the attribute of resource extremeness. They should give precedence to sites with lower extremeness or to locations where mitigation measures, such as large-scale, long-duration energy storage, are technically feasible and economically justifiable.

Renewable energy quality trilemma and relative prioritization of quality attributes across variable renewable penetration levels.

Full size image

Wind and solar droughts pose serious risks to systems relying on renewable resources; identifying and characterizing these threats can provide essential information for achieving power system reliability. We quantify the intensity and duration of extremes and develop a statistical framework to identify three categories (i.e., blue, orange, and red) of wind and solar droughts and investigate their spatial and temporal coincidence. We include three indicators (resource availability, variability, and extremeness) that, together, comprehensively assess renewable energy quality. Expanding on existing research, such as Antonini et al.²⁴, our work uniquely considers both wind and solar resources simultaneously. Our findings, which build on the analysis of the spatial and temporal coincidence of these two resource types, contribute novel insights valuable for the strategic deployment of decarbonized energy systems.

This study contributes and extends our understanding of renewable energy quality. Existing studies, which solely focus on the prevailing metric of resource availability, fail to capture the importance of variability and extremeness for renewable energy deployment, and may result in inadequate electricity infrastructure deployment and risks to power system reliability. Lack of sites that excel in all three quality attributes presents a trilemma to investors, system planners, and policymakers. This trilemma of renewable energy quality can help conceptualize three energy policy objectives, which are often difficult to reconcile and in conflict with each other. Stakeholders must prioritize one of the three attributes at different penetrations of variable renewables: at low penetration levels of variable renewables, resource availability will need to be prioritized; as the penetration of variable renewable energy increases, system planners must prioritize the attribute of resource variability at higher penetration levels; and at deeply decarbonized systems, planners will need to focus on the attribute of resource extremeness. The three dimensions of renewable energy quality that are offered here yield new insights for designing and developing policies to assist regions in understanding the progress of their energy transition and potential obstacles to decarbonization.

In China and many other countries, the resource availability has traditionally been the decisive metric for renewable energy project development. National governments and consulting companies publish resource availability maps to guide these projects. Over time, system planners have recognized the variability of these resources. Consequently, Chinese local governments now require solar and wind power plants to be equipped with energy storage solutions. The minimum duration for storage is set at 1–4 h, with a minimum power capacity of 10–30% of the solar or wind power capacity, depending on the location. This study proposes a third metric, extremeness, which suggests that project developers should not only install short-duration energy storage but also consider long-duration energy storage technologies or alternative solutions. This recommendation is based on our map of extremeness, alongside the existing resource availability maps. Additionally, our findings on the spatial and temporal co-occurrence of energy droughts indicate that some regions could share their long-duration energy storage. If portable, these storage solutions could be rented and transported to specific sites during certain seasons. Also, regions where the droughts of solar and wind are not likely to coincide, could balance their approach with both resources.

Wind and solar droughts present different temporal patterns, but coincide at noon and afternoon in spring and winter. The red droughts, the category representing the highest intensity and longest duration droughts, are more prevalent in these periods. Summer has no red coincident droughts. Planners in summer-peaking power systems might be concerned that a system dominated by renewables could exhibit reliability problems in summer, due to high demand. However, our findings indicate that extreme low electricity generation events are unlikely to happen in summer. In comparison, spring and winter could pose serious challenges to power system reliability, because most renewable droughts occur in these seasons and electricity demand is also high.

Renewable energy droughts cluster in areas which generally have higher resource availability. The most severe (red) droughts are more centered in these areas. More importantly, these areas show strong spatial dependences of renewable energy droughts, posing risks to geographically expansive power grids. This highlights the contribution of our work to the selection of suitable sites for renewable energy development, even in provinces that are broadly considered resource-rich. Existing power system planning studies have demonstrated the large-scale deployment feasibility of renewable facilities in areas which have high resource availability, but they invariably overlook the history and potential repetition or exacerbation of renewable energy droughts and their effect on the operational reliability of power systems. Power system planners and operators must consider the possibly disastrous consequences of spatially concurrent renewable energy droughts when deploying renewable facilities. Also, the reliable operation of future power systems with higher renewable energy penetration will require a robust and interconnected power grid with increased transmission capacity, flexible resource sharing at the national level, and a national integrated dispatch mechanism.

While this study has exposed that at many sites in China, the availability, variability, and extremeness of renewable energy tend to manifest in undesirable directions, similar phenomena may occur at other locations worldwide. However, the underlying meteorological and geographical mechanisms driving this trilemma remain unclear and demand further investigation. Prior research has indicated a correlation between renewable energy droughts and the presence of extensive high-pressure systems over central Europe^49,50. We hope this study will inspire meteorologists and climate scientists to delve deeper into the mechanisms underlying this phenomenon.

The energy system is profoundly affected by climate change⁵¹. While most research focuses on the effects on generation or resource potential^51,52,53,54, the impact on the variability and extremeness of local resources is less understood. Recent studies have analyzed how historical climate change has affected long-duration energy supply shortages⁵⁵, but they do not provide guidance for system planners on where to install renewable energy infrastructure to mitigate these impacts. Although our research does not project how future climate change may affect resource quality in three dimensions, it lays a strong foundation for future studies. It can help answer critical questions such as where to install renewable infrastructures to adapt to future climate conditions.

As the world transitions to decarbonized energy systems over the coming decades, the methods developed for China can be employed in other countries to identify renewable energy droughts and comprehensively assess renewable energy quality. The three categories proposed in this study can be used as a reference to formulate standard metrics for renewable energy droughts helpful to monitor and alert of potential power shortages. The results provide useful information in the strategic planning of renewable energy, energy storage (especially long-duration energy storage), transmission systems, reserve margins requirement, and dispatch mechanism for policymakers, system planners, and investors. Our work is essential to inform strategies for a reliable decarbonized energy system and form policies to insulate society from energy disruption, while eventually contributing to the reliability and sustainability of low-carbon future power systems.

Data

We analyze 19 years of time-series data of wind and solar resources, at high spatiotemporal resolution, to assess renewable energy quality in China. In each 90-m × 90-m grid-cell, we select eligible sites for wind turbine and solar panel deployment. These eligible sites satisfy criteria for wind turbines and solar panels siting as indicated by high-resolution data on land use, surface elevation, and geomorphology, as well as wind speeds and solar irradiance. We then aggregate these sites into grid-cells of 0.5-latitude-degree and 0.625-longitude-degree and use the corresponding data on wind speed and solar irradiance during each hour of the period 2000–2018 to estimate their hourly wind and solar energy capacity factor (CF). The hourly CF (in the range of 0–1) refers to the proportion of electricity produced relative to the generation that would be obtained if the wind turbine or solar panel operated at 100% nameplate capacity in each hour. Wind speed and solar irradiance information were taken from the NASA’s MERRA-2 (Modern-Era Retrospective analysis for Research and Applications-2) dataset⁵⁶. The detailed methods used for identifying sites suitable for wind turbine and solar panel deployment and estimating the hourly CF value of wind and solar energy in these suitable sites can be found in Li et al.⁵⁷ and are also provided in the Supplementary Information (SI), S1.

Defining renewable energy droughts and characterizing their severity

We propose a statistical method to identify and characterize renewable energy droughts. Renewable energy droughts are defined as episodes when electricity generation is significantly lower than historical averages, i.e., renewable energy droughts are abnormal deviations from expected values. For example, we do not label an event of low solar energy generation during the nighttime as an extreme event, because, although its electricity generation is extremely low (i.e., zero), this is expected. Therefore, the identification of episodes of renewable energy droughts requires comparing the actual generation with its historical average.

We identify renewable energy droughts using hourly CF values to measure two drought characteristics that determine the risk posed to power systems: intensity and duration. We define three categories of renewable energy droughts to represent the potential risk posed to the power system: blue, orange, and red. These three categories represent increases in the intensity and duration of the droughts and are calculated as a function of the difference between the expected value of the CF—estimated as the historical average CF—and the actual CF value. The categories of droughts are also a function of the magnitude of the discrepancy between the expected and observed CF measured in comparison with the percentiles of the historical distribution. Consistent with previous studies^58,59, the blue droughts are defined as episodes when the actual CF value minus the expected CF value for 3 consecutive hours is below the 10th percentile; the orange droughts occur when the actual CF value minus the expected CF value for 3 consecutive hours falls below the 5th percentile; and episodes, when the actual CF value minus the expected CF value for 6 consecutive hours falls below the 1st percentile, are called red droughts.

Here, we use the long-term (i.e., 19-year) average CF during a specific period as an estimate of the expected CF during this period of the year. We divide the 8760 h of a year into 96 typical periods of 3-h duration and 6-h duration, each corresponding to one of the 4 seasons and 24 daily periods. The four seasons in this research refer to spring (March to May), summer (June to August), autumn (September to November), and winter (December to February). The 24 periods of 3-h duration are 01:00-03:59, 02:00-04:59, 03:00-05:59,…, 22:00-24:59, 23:00-01:59, 24:00-02:59; and the 24 periods of 6-h duration are 01:00-06:59, 02:00-07:59, 03:00-08:59,…, 22:00-03:59, 23:00-04:59, 24:00-05:59. The long-term average CF observed over these 96 periods is taken as the best estimate of the expected renewable generation during the given periods, so we define a deviation as the actual CF value minus the expected CF. Large negative deviations indicate greater drought severity.

To define renewable energy droughts, we follow four consecutive steps, described below for blue droughts as an example; the other two categories of droughts can be defined similarly.

First step: Calculating the actual capacity factor (CF) value for 3 consecutive hours

We take the 19 year-long hourly time-series data for a given site i and partition these hours into 3-h-long segments on a rolling basis. Each of these time segments is indexed by t (t ranges from 1 to T, T is the total number of hours in the 19 years and T = 166,560 (The 19 years include 5 leap years with an extra 24 h each. For simplicity, throughout the manuscript we use 8760 to represent 8760/8784 h of a year in the research period.)) and each of the hours is indexed by h (h ranges from 1 to T+2). In this way, segment t = 1 includes h = 01, 02, 03, t = 2 includes h = 02, 03, 04,…, t=T includes h = T,T + 1,T + 2. The final time segment includes the last hour in 2018 and two additional hours in 2019. Then, for each hour h, we calculate the hourly CF value ({{CF}}_{i,h}.) To find the three-hour actual CF for segment t, ({{CF}}_{i,t}), we average the hourly CF values for the 3 consecutive hours in time segment t as shown in Eq. 1. This is repeated for all sites i, and for all time segments t:

({{CF}}_{i,h}) is the hourly CF value at site i in hour h, ({{CF}}_{i,t}) is the average of CF values at site i for the three consecutive hours in time segment t.

Second step: Calculating the expected capacity factor (CF) value for 3 consecutive hours

We assign each time-segment t to one of the 96 typical three-hour periods (indexed by s, s ranges from 1 to 96) according to the time of the day and season. For example, time segments starting at hours 1, 25, 49, 77… all belong to typical period s = 1 because they begin at the first hour of a day in the first season (i.e., in winter). We denote the number of time segments t that belong to a typical period s during the 2000–2018 years as ({{NS}}_{s}) (a number around T⁄96 or 1700 depending on the season). We calculate the expected value of the CF in a typical period s, ({{CF}}_{i,s}), as the average across all the ({{NS}}_{s}) time segments t included in that typical period s for all the years in the dataset. This is repeated for each site and each of the 96 typical periods:

({{CF}}_{i,s}) is the average of the three-hour observed CF value at site i at each time segment t belonging to typical period s; ({{NS}}_{s}) is the total number of time segments t that belong to a typical period s. ({{CF}}_{i,s}) can be taken as the expected CF value during this typical three-hour period for this site.

Third step: Calculating the difference between the actual and expected capacity factor (CF)

For each time segment t, we calculate the difference between the actual CF value and its expected value for the corresponding typical period s:

(Delta {{CF}}_{i,t}) reflects the deviation of observed CF from its expectation (given historical data for this site); the more negative the value of (Delta {{CF}}_{i,t}) the more severe the drought.

Fourth step: Defining renewable energy droughts at a site i and province p

Given that each time segment t comprises 3 h, we label episodes with ∆CF below the 3.3th percentile as blue droughts. In this way, no more than 10% of total hours in the historical period analyzed are labeled as blue droughts. This is expressed as

Where F is the Cumulative Probability Distribution (CDF) of ∆CF, (Pleft(Delta {CF}le {z}_{alpha }right)) is the probability (in a frequentist sense) of ∆CF being less than ({z}_{alpha }), α is the 3.3th percentile; ({z}_{alpha }) is the 3.3 percentile value of ∆CF.

If the (Delta {{CF}}_{i,t}) value falls below the 3.3th percentile, we label this episode of low renewable generation at site i for time segment t as a blue drought, and assign a value of 1 to the segment blue-drought indicator variable ({k}_{i,t}). This is expressed as

({k}_{i,t}) is a binary variable that indicates whether there is a blue drought at site i during time segment t. To facilitate the analysis of spatial and temporal distribution of droughts and determine the coincidence of wind and solar shortages, the drought designation must be applied to all the hours in the segment, t. Hence we define a binary variable ({g}_{i,h}), that indicates whether there is a blue drought at site i during any time segment t that contains hour h. For example, if time segment t, which includes hours t, t+1, and t+2, experienced a blue drought, then all hours in the segment are marked as having a blue drought, i.e., ({g}_{i,h=t}={g}_{i,h=t+1}={g}_{i,h=t+2}=1).

We calculate the prevalence of droughts for all sites in a specific province and hour as the proportion of sites with droughts in this province. It is expressed as

({g}_{p,h}) reflects the prevalence of droughts occurring in province p at hour h; it is a continuous variable in the range of 0–1, with zero indicating no droughts in any site and 1 indicating droughts in all sites. ({{NP}}_{p}) is the number of eligible sites for wind turbine or solar panel deployment in province p.

In order to quantify a site’s propensity to exhibit renewable energy droughts, we propose the spatial Relative Drought Frequency (RDF) index, which is defined for a specific site (and a specific resource, either solar or wind) as the ratio of the proportion of droughts observed in this site relative to the average proportion of droughts observed in all sites. The RDF at site i is expressed as

where NI is the number of total eligible sites for wind turbine or solar panel deployment in China. ({{RDF}}_{i} > 1) indicates that site i is more likely to experience a resource drought than the average site in the country; while ({{RDF}}_{i} < 1) indicates that in this site a renewable energy drought is less likely to occur compared to the average site. By comparing the RDF for various sites, we investigate the spatial pattern of renewable energy droughts.

Similarly, we propose the temporal RDF index, which is defined for each hour as the proportion of droughts observed during this particular hour, relative to the average proportion of droughts observed on all hours. The RDF at hour h is expressed as

We use the spatial and temporal RDF indices to measure the extremeness of renewable energy resources in various sites and times.

Definition of temporal coincident wind-solar energy droughts

We define and identify episodes when wind and solar energy droughts co-occur (called coincident wind-solar energy droughts) in a similar way to that presented in Section 4.2. for droughts of just one resource, with the difference being that we use the sum of wind and solar capacity factors, and we define coincident wind-solar energy droughts for provinces but not sites (because not all areas can host both wind and solar energy facilities).

Availability

We use the long-term average of annual CF values from 2000 to 2018 to represent the resource availability of a site. Annual CF (in the range of 0–1) is widely used in assessments of renewable energy resource quality, because it is a measure of the potential to convert solar irradiance or wind speeds into electricity. All things equal, a higher CF indicates higher electricity generation and relatively lower levelized cost of electricity (LCOE) for the analyzed site. We calculate annual CF values from hourly CF values. The sum of hourly CF values (a number in the range 0–8760) divided by 8760 equals the annual CF value.

The long-term average of annual CF at site i (({{CF}}_{i})) is expressed as

where ({{CF}}_{i,y}) is the annual CF value at site i in year y. Y is equal to 19, the number of years covered by the data.

Variability

We average the annual CF standard deviations over the 19-year period to represent a site’s resources variability. Standard deviation (in the range of 0–1) is a useful indicator of statistical dispersion, and a higher standard deviation value indicates greater variability of renewable energy.

The long-term average of the annual CF standard deviation at site i (({delta }_{i})) is expressed as

where ({delta }_{i,y}) is the standard deviation at site i in year y.

Spatial dependence of renewable energy droughts

Spatial dependence between each pair of sites

We use the Jaccard similarity coefficient to represent the spatial concurrence of renewable energy droughts between two sites. The Jaccard similarity coefficient of renewable energy droughts between sites i and j is expressed as

({A}_{i}cap {A}_{j}) is the number of hours with concurrent droughts in sites i and j, while ({A}_{i}cup {A}_{j}) is the total number of hours with droughts in either site/both sites. (0le {J}_{i,j}le 1); the higher the ({J}_{i,j}) value, the higher the concurrence of droughts in sites i and j. ({J}_{i,j}=1) indicates that (left|{A}_{i}cap {A}_{j}right|) equals (left|{A}_{i}cup {A}_{j}right|) and renewable energy droughts in sites i and j always occur simultaneously; ({J}_{i,j}=0) indicates no concurrent droughts in sites i and j.

We link the sites by the Jaccard similarity coefficient between each pair of sites employing the co-occurrence network analysis. Co-occurrence networks are graphical representations of how frequently variables appear together and widely used to reflect the joint spatial effects. We use this method to depict the spatial dependence and coincidence of renewable energy droughts among sites.

Spatial dependence between each pair of provinces

We use the Pearson correlation coefficient to measure the spatial correlation of renewable energy droughts between two provinces. The Pearson correlation coefficient ρ quantifies the strength and direction of the linear relationship between the droughts occurrence in provinces ({p}_{a}) and ({p}_{b}). We define m as the set of hours where either ({p}_{a}) or ({p}_{b}) had a drought, and M the total number of hours when either province had a drought. In the calculation of the Pearson correlation coefficient, we only consider the hours in the set m (i.e., hours when at least one province had droughts) and exclude the hours without any droughts. This is because droughts are rare events and considering the hours without any droughts would result in an artificially high correlation coefficient. This would negate the possibility of investigating the spatial coincidence of wind and/or solar droughts.

(-1 , le , {rho }_{{p}_{a}{p}_{b}}le 1); ({rho }_{{p}_{a}{p}_{b}} > 0) indicates that provinces ({p}_{a}) and ({p}_{b}) are positively correlated in terms of the prevalence of renewable energy droughts; otherwise, they are negatively correlated.

We cluster the provinces by the Pearson correlation coefficient between each pair of provinces employing the hierarchical clustering method. Hierarchical clustering is an algorithm widely used to group a set of objects in a way that objects in the same cluster are more similar to each other than to those in other clusters. We use this method to depict the spatial dependence of renewable droughts among provinces.