Photovoltaic (PV) installations have rapidly and extensively been deployed worldwide as a promising alternative renewable energy source. However, weather anomalies could expose them to challenges in supply security by causing very low power production. Using reanalysis weather data from 1986 to 2021 and a high-resolution global inventory of PV installations, we assess the impact of extreme low-production (ELP) events across various regions. Our results reveal that regions between 60°N and 60°S experience an average of 27 ELP events annually, with 17% of these events being high-intensity. Regions with dense PV installations—including Southern China, Central and Northern Europe, Central and Eastern America, and Japan—are particularly affected. These areas, which collectively host approximately half of the global PV installations, see 44% of ELP events being high-intensity. Maintaining a daily backup supply equivalent to the average event intensity could recover 39% to 81% of events across different sites. This strategy helps ensure a stable energy supply despite the unpredictability of extreme weather events.
Introduction
Driven by technological advances, falling costs, and a growing commitment to sustainable energy, photovoltaic (PV) infrastructure is expanding rapidly across the globe1. At the end of 2022, the installed PV capacity worldwide reached about 1.2TW2. According to IEA, the world is on course to expand its renewable capacity, and the global power mix is expected to be transformed over the next 5 years3. For most countries, it is a critical step toward achieving carbon neutrality by transitioning to a low-carbon energy source4,5,6,7. However, it has been proved that PV power easily exhibits extreme low production (ELP) owing to various weather-related factors or weather pattern8,9,10. For example, the efficiency of PV panels significantly drops during extreme heat. Cloud and aerosols result in very low power production by reducing near-surface solar radiation11. In Europe, it was found that all regions have experienced periods of very low solar power over the past 23 years (1995–2017), though the severity and driven weather patterns differ12. As reliance on PV power grows, ELP events can cause prolonged or severe power shortages13,14, threatening power supply security and leading to significant social and economic costs15.
There has been widespread recognition of the importance of incorporating the intermittency caused by extreme conditions into planning and management practices16,17. Short-term forecasting of PV power generation based on weather predictions helps in preparing for immediate challenges18,19,20, while long-term risk assessments inform strategies for mitigating the impact of extreme events, which is especially useful when extreme weather is difficult to predict21,22,23. Great attention has been paid to global intermittency resulting from weather24,25,26 and its impact on power supply reliability26,27,28,29. However, gaps remain when developing suitable solutions to mitigate the risk of extremely low production for the widely deployed PV infrastructures. Current approaches often rely on spatial distribution assumptions that do not fully reflect real-world deployments27,29,30, leaving the underlying spatial expansion and exposure patterns of the PV infrastructures still unclear. Furthermore, while high-resolution data has been used to assess the impact of extreme weather events on renewable energy system13,14,28,31,32,33,34, these studies are often limited to specific geographic areas due to computational cost or data availability constraints, resulting in a limited breadth of available information.
This study provides a global perspective on the characteristics of ELP events and their impact on widely deployed PV installations. We do so by first estimating the daily PV power generation during 1986–2021 using ERA5 reanalysis weather data. Utilizing a common approach for assessing extreme event probabilities, we identify and classify ELP events into nine types based on intensity and duration. We analyze event frequency and the total number of low-production days for each type, and map these findings to the locations of PV facilities, assessing their exposure. Furthermore, we explore potential measures to mitigate the effects of these ELP events. When applied in suitable regions, the solution offers the great potential to enhance PV power supply security. Our findings contribute to a better understanding of the risks posed by ELP events and underscore the need for effective adaptation measures to ensure PV power supply security.
Results
Global characterization of extreme low-production events
ELP events are identified using a percentile-based approach22,35,36, where the values falling below a specified percentile are classified as extreme (Methods). Specifically, we consider this threshold to be the 10th percentile of daily power generation values, indicating that an ELP day is one where power output is lower than 90% of all days in the observation period. An ELP event consists of a sequence of consecutive days where production is consistently below the 10th percentile. We categorize events based on intensity and duration, classifying them as low (5th–10th percentile), medium (<5th percentile), or high intensity (near-zero power) of short (<3 days), medium (3–7 days), or long (>7 days) duration, as illustrated in Fig. 1, and detailed in Supplementary Fig. 1. Events primarily characterized by longer durations (e.g., Category 3) can lead to the continuous consumption of power storage without replenishment, while events characterized primarily by intensity (e.g., Category 7) may cause immediate, large-scale disruptions to power supply.
From 1986 to 2021, we observe an average of 27 ELP events annually (range: 14–66 times) across the global 0.25° × 0.25° grid, with an average of 56 days occurring annually (range: 32–219 days). Short-duration events occur most frequently, comprising 89.5% of all events and 52.8% of the total ELP days. Areas with higher PV power generation potential, characterized by ample solar radiation and clear sky, tend to experience low or medium-intensity events more frequently, whereas areas with poorer PV power generation potential show a higher occurrence of high-intensity events (Figs. 1 and 2). For instance, areas with abundant solar radiation resources, such as Africa, the Arabian Peninsula, southern Asia, western and northern China, Mongolia, northeastern South America, and the western United States, exhibit lower intensity and shorter duration of ELP events and consequently enjoy higher power security. Conversely, areas with relatively poor PV power generation potential, including southeastern China, central and northern Europe, the eastern United States, and high-latitude regions in the Northern Hemisphere, are more prone to high-intensity events that may persist for days.
The seasonality of high-intensity events is clearly obvious. In the Northern Hemisphere, particularly in southeastern China, the eastern United States, and Europe, approximately 70% of high-intensity events occur from November to February. Moreover, 84% of high-intensity events of long duration are concentrated north of 66°N, attributed to the polar night phenomenon. To eliminate effects related to seasonal variation, we further exclude the events where the threshold of the 10th percentile is already exhibiting a high-intensity state. As shown in Supplementary Fig. 2, in these areas, high-intensity events resulting from unstable weather occur an average of 1–18 times annually, corresponding to 1–23 days. High-intensity event of short duration is the main category.
Notably, even in regions abundant in solar radiation resources, we still observe some events showing medium intensity lasting 3 days or more (categories 5 and 6) in northern and southwestern Africa, the Arabian Peninsula, South Asia, western and southeastern China, Brazil, western America, and northern Australia, with an average of 1–4 events annually, corresponding to 3–15 days. Such events are likely to have a stronger impact as they endure longer.
Exposure of global PV installations to extreme low-production events
The existing PV installations37 are predominantly situated in areas highly exposed to poor PV power generation potential and consequently significantly affected by weather variability (Supplementary Fig. 3). To assess the impact, we analyze the average intensity of these events experienced by PV installations at different PV power generation potential levels during 1986–2021 and compare it with the global explorable land mass average within the range of 60°N–60°S (Fig. 2). Globally, PV installations experience an average event intensity (the percentage of lower-than-10th-percentile levels) of 54%, which is 13% higher than the global land mass average. 66% of PV installations are situated in areas with daily PV power generation potential of 0.2–0.3?kWh/m2, approximately corresponding to the 20th–60th percentiles of the PV power generation potential across the global explorable land mass. These PV installations in such areas experienced severely high event intensities, with an average intensity of 73%, equivalent to 1.8 times the global average. Japan and China, where the event intensity is stronger (averaging 80%), host nearly half of them. In addition, 13.9% of PV installations are situated in areas with daily PV power generation potential lower than 0.2?kWh/m2, primarily in Germany, the Czech Republic, the United Kingdom, and China, with an average daily intensity of 88%. 20.6% of the PV installations are situated in areas with higher PV power generation potential (>0.3?kWh/m2), including China, the United States, Turkey, Spain, and Pakistan, with an average daily intensity of 40%.
As shown in Fig. 3a, global PV installations face a higher risk of high-intensity events compared to the global land average between 60°N and 60°S. Specifically, 75.8% of these PV installations experience more frequent high-intensity events than global land averages. On average, PV installations experience high-intensity events 13 times annually, compared to 5 times for the global land average. Similar results can be observed after removing the seasonal high-intensity days (Supplementary Fig. 4), with 82.3% of PV installations experiencing more frequent high-intensity events than the global land average.
To enhance the visualization of the spatial distribution, we have determined the primary category for each PV site based on the highest frequency (Fig. 3b). Approximately 56% of PV installations are predominantly affected by high-intensity events of short duration in terms of event frequency. These installations are concentrated in central and northern Europe, southern China, Japan, and the eastern United States. Besides, 5% of PV installations are affected by high-intensity events of medium duration in western China, western United States, and southern Africa. Although the frequency of longer-duration events may not be the highest in these regions, 30% of PV installations are primarily affected by high-intensity events of long duration in terms of the total number of ELP days (Supplementary Fig. 5).
Small-scale PV installations are prevalent, with the installed capacity between 10?kW and 10?MW accounting for around 90% of global PV installations and 30% of global installed capacity. Of these, 51% are “commercial and industrial” installations (10?kW?1?MW)37,38, and 49% are small utility-scale installations (1–10?MW). These installations are often sited near human settlements and are convenient to meet local power demand, however, they face a significantly higher risk of ELP events. Specifically, they experience roughly 3.9 times more high-intensity events compared to installations in excess of 100?MW and about 2.6 times more than those in excess of 50?MW.
The potential of maintaining a daily backup supply to mitigate extreme low-production events
To examine possible measures for mitigating the effects of ELP events, we evaluate the viability of a simple strategy: maintain a daily backup supply for each site. This approach enables the swift deployment of alternative energy sources, such as emergency oil-fired generators, as dispatchable resources to mitigate the impact of power shortages during ELP events. Specifically, we use the average intensity of the local ELP event (calculated as the absolute reduction relative to the 10th percentile) as the daily backup supply level and recalculate the distribution of the ELP event. In addition, we exclude seasonal variation effects, meaning that PV should not be expected to act as a stable power source during sustained downturns. For most current PV installations, this backup level corresponds to 10–30% of the long-term average daily power generation potential, depending on the ELP risk level.
Our analysis indicates that this strategy demonstrates considerable potential for adaptation, resulting in a substantial reduction in the frequency of each category of events (Fig. 4b). It is projected to decrease all types of events by 39–81% across various sites, with reductions of 84% of medium-intensity events lasting 3 days or more (categories 5 and 6).
While a higher backup supply can certainly mitigate ELP events, they also incur additional costs and lower utilization. To assess the effects of different backup supply levels on adaptation to ELP events, we calculated the reduction in the frequency of ELP events at various backup supply levels (0.5×–1.5× the average daily intensity of ELP events). As backup supply levels increase, high-intensity or long-duration ELP events are converted to low-intensity events of shorter duration, while low-intensity events of short duration are gradually eliminated (Supplementary Figs. 6–10). Globally, when the backup supply level is halved (0.5×), the frequency of short-duration and mild events increases by 17%, while moderate events decrease by 21–74%. When backup supply levels reach approximately 0.8x and above, the frequency of all events will decline (Fig. 4b).
We observe that the average daily intensity is quite high in specific high-risk areas, exceeding 80% of the 10th percentile. This implies that managing ELP events at particular risk levels demands nearly equivalent backup supply levels, making it highly improbable to effectively eliminate most ELP events with a lower backup supply level (Fig. 4a). Conversely, regions with high PV power generation potential and predominantly affected by mild or short-duration events can effectively mitigate ELP events with a lower backup supply level. For instance, in North Africa, West Africa, and South Asia, a backup supply level of 0.5× (equivalent to 12% of the 10th percentile on average) can notably decrease ELP events by over 35–40%. Elevating the backup supply level to 1.5× (equivalent to 37% of the 10th percentile on average) could further reduce ELP events by 70–72%. However, for nearly half of the global PV installations concentrated in China, Japan, Germany, South Korea, the UK, and the US, backup supply levels at 0.5x have already reached an average of 70% of the 10th percentile, enabling the recovery of only 28% of ELP events at this level. In general, comparable backup supply levels have aided in effectively mitigating long-duration or high-intensity events by transforming them into milder or shorter-duration events. However, addressing the resultant increase in low-intensity and short-duration events entails additional costs, necessitating further increases in backup supply or the implementation of other adaptation strategies.
Discussion and conclusion
This study assesses the impact of ELP events on PV power supply security across different regions, offering a global perspective incorporating the distribution of current PV installations. Generally, the spatial distribution characteristics of ELP events demonstrate a strong correlation with local annual average PV power generation potential. Consistent with previous research on the influence of solar intermittency on PV reliability26, we observe that regions with favorable PV power generation potential tend to experience fewer effects from ELP events. However, the current distribution of PV installations globally does not align optimally with these areas, possibly due to challenges in construction, sparse populations, or limited economic development. Consequently, many PV facilities are situated in regions such as southern China, Europe, Japan, and the eastern United States, characterized by higher ELP risk and frequent high-intensity events. Comparatively, sites with PV installations demonstrate a 2.7 times higher frequency of high-intensity events than the global land average. Notably, numerous small-scale PV installations near human settlements—“commercial and industrial” installations and some utility-scale installations with small installed capacity—are facing heightened risks associated with ELP events. It is important to note that our analysis does not include residential PV systems, which could also be subject to similar risks. Future studies could benefit from incorporating residential PV data to provide a more comprehensive assessment of the risks associated with ELP events.
Our results highlight the critical need to prepare for ELP events, which present greater challenges than conventional power gaps. Characterized by prolonged durations or high intensities, ELP events can lead to power shortages that exceed the capacity of conventional reserves, especially in some regions with dense PV installations where high-intensity events of long duration have occurred.
One potential solution for reducing ELP events is to maintain a daily backup supply for each site. It enables the rapid deployment of alternative resources through various strategies, such as the use of emergency fuel generators. Our findings suggest that the sites with abundant solar resources, where low-intensity events are the primary disruptors of PV power generation, can effectively recover most extremely low production days with a low backup supply level. In contrast, for high-risk PV sites, high-intensity events need to be gradually mitigated to lower-intensity events, followed by an increase in backup supply levels to recover these additional low-intensity events.
While our proposed strategy is only one of several possible solutions, it helps quantify the energy storage needed to effectively manage ELP events. Diversifying energy sources—such as incorporating wind and hydropower—may also effectively mitigate ELP events. Besides, transmission infrastructure and on-grid connections39 offer potential solutions for reducing ELP events, especially for countries with numerous dispersed PV installations, as the likelihood of extreme weather simultaneously impacting multiple regions is significantly lower. For example, in countries like China and the United States, regional grid connections can alleviate the majority of intermittent effects27,30. When the benefits of regional grid connection in these nations are paired with backup supply, it may become feasible to effectively balance emergency backup supply levels with dispatch-dependent infrastructure. These assessments should be combined with electricity demand data to further evaluate regional benefits in the future. Finally, climate change is likely to be a prominent influencing factor in the future. Further research into climate change impact and multiple solutions will be valuable in addressing these complex challenges in the future.
Despite the progress made in this study, there are still uncertainties that should be acknowledged. Several factors can affect the estimation of PV power generation, including panel tilt, azimuth, and system efficiency. For instance, some new PV plants have started to use bifacial PV modules to improve energy capture by additionally using reflected sunlight from the ground40,41, which can help reduce the risk of ELP events. Additionally, using high-efficiency materials or designing systems to perform better in high temperatures42,43 could further reduce the impact of ELP events. However, in regions prone to heavy soiling, such as heavily air polluted and desert regions with low precipitation, the risk of ELP events might be higher than expected due to the PV generation reduction44. Despite these uncertainties, our results findings provide early warning and implications for energy stakeholders and future research. Through examining ELP events at specific exceedance probability levels for each local region, this study highlights that ELP events pose a relatively high risk to global PV installations, especially in areas with dense deployment, and it is not expected to be altered.
In conclusion, it is crucial to comprehend the impact of ELP events in order to establish safe renewable energy alternatives during the transition to sustainable energy. Mitigation measures are necessary for most current PV sites to ensure the seamless integration of a stable alternative energy source into the electricity supply system. However, implementing such measures can be costly, often involving expanding infrastructure, upgrading technology, and requiring long service periods. At the regional level, conducting detailed cost–benefit analyses are crucial to ensure the effectiveness of energy-generating portfolios. This study provides valuable scientific insights for policymakers and energy planners, promoting more appropriate siting or adapting appropriate supporting measures to minimize disruptions and optimize their effectiveness.
Methods
Estimating PV power generation based on the PVLIB solar PV system model
Global PV power generation is estimated based on the PVLIB model, which was developed by Sandia National Laboratories45. PVLIB provides a set of functions and classes for simulating the performance of PV energy systems at a given time and location. Generally, the solar radiation reaching the ground is first converted into effective radiation on the PV panel according to the azimuth and tilt angle. The effective radiation is then converted into direct current (DC) power generation by the PV module and finally converted into alternating current (AC) power generation by the inverter.
The required weather data includes solar radiation, air temperature, and wind speed. Solar radiation determines how much electricity can be converted. A high ambient temperature is considered to work against the efficiency of a PV panel, while wind can facilitate heat dissipation and cooling of a panel46. Considering that the estimation of sun-related parameters depends on the input of specific time and location information, all input weather data are separated by the hour to reduce uncertainty. Tilt angle and azimuth are the most critical parameters in PV panel configuration because they directly affect the amount of solar radiation a PV panel receives. For the azimuth, we assumed that all panels faced due south/north (in the Northern/Southern Hemisphere), which is generally considered an optimal azimuth approximation and has been used in many related studies24,27. For the tilt angle, we placed the PV panels at a fixed tilt angle and used an optimal tilt angle for each location to maximize solar radiation capture. We applied an improved third-order polynomial model, which depends on latitude, to determine the optimal tilt angle47. This method performs better in regions above 40°N latitude compared to previous linear fits. On a global average, the performance of this fixed tilt configuration is broadly comparable to the radiation received by single-axis vertical tracking systems (which swivel vertically around a horizontal axis), and it is approximately 13% lower than that of single-axis horizontal tracking systems (which swivel horizontally around a vertical axis) and 17% lower than two-axis tracking systems.
The Sandia PV Array Performance Model is applied to convert the effective irradiance to DC power. The current–voltage curve describes the module production, which varies as a function of irradiance, panel temperature, and cell material. Empirical coefficients are provided in the Sandia database based on laboratory experiments. Here, we selected the “Canadian_Solar_CS5P_220M” crystalline silicon PV module in the Sandia database, with a maximum power of 220?W under reference conditions. We assumed that the inverter could perfectly track the maximum power point. Then, the “PVP2500 240?V” inverter with a 2500?W AC power rating is used to convert DC power to AC power in this study.
The production results are then converted into the average power generation, which is the product of power and time. Two additional coefficients are set and multiplied by the power generation. The first is the available area coefficient, and the second is the system efficiency loss coefficient. Considering the actual need to provide a channel between the facility and the PV modules for maintenance, the available area coefficient is set as 0.9, i.e., the ratio of the effective area of the module to the required land area48. Several factors can affect the system performance and reduce the actual PV power generation, such as shading and soiling, wiring loss, instability of PV conversion efficiency, and system failure and maintenance. To simplify the simulation, a system efficiency of 80% is used uniformly in this study49,50. Following the test module provided in the Sandia database, all results are finally converted into a sequence of hourly power generation per unit area during 1986–2021.
Global weather data
All global weather data are obtained from the ERA5 reanalysis dataset, which provides a large amount of hourly atmospheric and land surface estimates available at 0.25°?×?0.25° resolution for public download51. We use surface solar radiation downward, i.e., the solar radiation reaching a horizontal plane at the surface with clouds and aerosols taken into consideration. To estimate the effective irradiance on the panel with a specific tilt angle, the direct horizontal and diffuse radiation must be further distinguished from the surface solar radiation downward. Here, the diffuse fraction is calculated using the Boland–Ridley–Lauret (BRL) model52, a multi-variable logistic model that considers the clearness index, solar elevation angle, persistence of sky conditions, and apparent solar time. In this study, we adopted the parameters estimated by Lauret et al.53 using the Bayesian parameter estimation method, which is based on nine locations with different environmental conditions in Europe, Africa, Australia, and Asia. In comparison with other local models, the BRL model is generic and more suitable for application in other locations worldwide. Additionally, air temperature at 2?m above the surface and wind speed at 10?m above the surface are used to estimate the impact of environmental conditions on PV panel performance.
Definition and analysis of extreme events
We define an ELP day as one where PV power generation falls below the 10th percentile, indicating the power generation on this day is lower than what is typically observed on 90% of days. To avoid strong variability on the daily scale, we applied a 15-day rolling window to the daily power generation data over a 36-year base period (1986–2021). This method is commonly used to analyze extreme events22, such as heatwaves35,36. Then, an ELP event period is defined as a contiguous sequence of days with extremely low PV production. Event intensity is defined as the percentage of lower-than-10th-percentile levels.
ELP events are classified into nine categories based on duration and intensity. Intensity is defined as the reduction from the 10th percentile. Three levels are set in terms of both duration and intensity. Specifically, the three intensity levels are mild (5th–10th percentile), moderate (<5th percentile), or severe (near-zero power) and the three intensity levels are short-duration (<3 days), moderate-duration (3–7 days), or long-duration (>7 days). The ‘near-zero power’ state refers to a condition where power generation is either zero or very low, producing either no power or only the minimal output required to meet the consumption demands of the PV system. Then, they are combined through a 3?×?3 matrix representing different ELP events, ranging from short-duration mild events with small impact to long-duration severe ones with a large impact (Supplementary Fig. 1). We calculated the annual frequency and days for each event category at each grid to determine the characteristics of ELP events in different regions. Complementary information could be considered to avoid one-sided information, e.g., to ensure a very long event is counted as only a single event.
The global inventory of PV installations used in this study is obtained from Kruitwagen et al37., who located PV installations globally based on remote sensing imagery and machine learning. This inventory includes 68,661 PV facilities with detailed geometries and features, focusing on installations larger than 10?kW. We extracted ELP event information for the centroid of each PV facility. Histograms are first used to identify the global distribution of each category of ELP events. Then, the major category with the highest frequency and greatest number of days is mapped.
Distinguishing exploitable areas
To better compare the risk at global PV installation sites with the global average, we implemented additional restrictions to exclude unexploitable areas, thereby avoiding their impact on overall risk calculations. The stringency of these constraints varies widely across studies30,49,54,55. Here, we imposed relaxed constraints to include more potential areas in our assessment. Specifically, areas were excluded if they met any of the following conditions: nature reserves, forests, water bodies, areas within a 150?m buffer of roads and railways, and slopes greater than 30°.
Nature reserves were identified using the World Database on Protected Areas (WDPA)56, the most comprehensive global spatial dataset available. The spatial locations of forests, water bodies, and urban areas were identified using the 2022 MODIS Land Cover Type Product (MCD12Q1)57 dataset at 500?m spatial resolution, utilizing the University of Maryland classification. Railway and road data were obtained from the OSM database58. Slope data were derived from the Global Multi-resolution Terrain Elevation Data (GMTED2010) at 500?m spatial resolution, developed by the U.S. Geological Survey (USGS) and the National Geospatial-Intelligence Agency (NGA)59.
All filters were initially applied at 500?m spatial resolution, and then the proportion of non-exploitable elements was calculated on a 0.25° spatial resolution map. Areas were considered exploitable if the proportion of non-exploitable elements was less than 50%.
Backup supply algorithm
When addressing the risk of ELP events in PV power supply, incorporating a daily energy backup is a concept that can be practically implemented. This involves prepositioning backup energy equipment, such as emergency generators or storage systems, at key locations to mitigate sudden drops in PV power generation.
For each target PV site or grid cell, we first gathered all power generation data for ELP days and calculated the absolute reduction in power generation relative to the 10th percentile for the corresponding date. To rule out seasonal blackout conditions, days with zero generation in the 10th percentile were excluded. The average intensity of these ELP days, after screening, was used as the reference value for the backup power level for each day. In this study, we set five different backup levels ranging from 0.5 to 1.5 times the reference value. After establishing the set backup power levels, we recalculated the power supply for each ELP day. Using the same method, we generated a new ELP event dataset post-backup deployment and reexamined the ELP event frequency for each category.
Data availability
ERA5 data is available at https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=overview. The parameters used to estimate PV module performance are available at https://github.com/NREL/SAM/tree/develop/deploy/libraries. The post-processing data used to produce the figures or statistical results in this paper is available at https://github.com/Vapson/Global_PV.git60.
