Analysis of photovoltaic power plant resilience under Iberian heatwaves Open Access

The ongoing climate change is producing significant impacts in all aspects of human life, with Europe at the epicentre of its impacts. Since 1979, Europe’s surface air temperatures have increased about three times as fast as the global average, in both winter and summer (Dong and Sutton, 2025). This excess warming is largely due to greenhouse gas emissions, changes in aerosols and shifts in atmospheric circulation patterns that bring more warm air into Europe (Dong and Sutton, 2025; Vautard et al., 2023). Regional projections confirm European land temperatures will continue to rise by 1.2–3.4°C under SSP1-2.6 and up to 8.5°C under SSP5-8.5 by 2100, with the strongest warming expected in northeastern Europe and inland areas of Mediterranean countries (EEA, 2025). As a result, the frequency, intensity and duration of extreme weather events, including heatwaves, floods, droughts, hailstorms, wildfires and high wind events, has increased (Vautard et al., 2023; Ruosteenoja et al., 2023). The 2022 summer was Europe’s hottest summer, triggering the most severe drought in 500 years and widespread wildfires across Southern Europe (IPCC, 2022). Future projections indicate higher risks of pluvial flooding in Northern and Western-Central Europe, while Southern and Western-Central regions will face more frequent droughts and extreme heat, with populations enduring over 60 days annually above 35°C by century’s end (IPCC, 2022).

Other hazards are also intensifying. Hail is already among the costliest weather-related events, with northern Italy recording up to 40 severe storms per 10,000 km² each year (EEA, 2021; Kahraman et al., 2024). Climate simulations suggest hailstorm tracks could expand by 15%–30%, with doubled frequency of > 50 mm hailstones (Brennan et al., 2025). Altogether, the financial burden of climate-related risks in Europe could reach €40bn annually by the 2080s, with €35bn attributable to heatwaves and droughts (IPCC, 2022). These changes already threaten energy infrastructure. Extreme events can reduce system reliability by up to 16% and grid integration capacity by 34% (Perera et al., 2020). Both centralised and distributed generation are vulnerable, with building-level PV systems especially exposed during prolonged heatwaves or islanded operation (Han et al., 2025). The economic losses amount to billions annually (Añel et al., 2024), with low-income households disproportionately affected by energy poverty and price volatility (Lei and Xu, 2024).

Solar photovoltaics (PVs) play a dual role in Europe’s energy transition: they are central to decarbonisation, with more than 339 GW installed by 2024 (IEA, 2025), yet are also highly exposed to climatic stressors. PV performance is sensitive to temperature, irradiance, aerosols, humidity and wind, many of which are shifting under climate change (Patt et al., 2013; Bamisile et al., 2025). Temperature is one of the most critical factors. Rising air temperatures reduce efficiency by 0.3%–0.5% per °C above 25°C (IEA, 2022), while heatwaves can drive module surface temperatures above 70°C, accelerating degradation (Islam et al., 2024; Omazic et al., 2019). Laboratory studies show crystalline silicon modules lose 0.46%–0.50% efficiency per °C, with heatwave conditions leading to 4%–10% efficiency drops between 35°C and 49°C (Razak et al., 2016; Hudișteanu et al., 2024). Thin-film technologies such as copper indium gallium selenide (CIGS) are less thermally sensitive, offering some advantages in very hot regions (Ebhota and Tabakov, 2022). Thermal stress also affects other system components. Expansion mismatches between polymers and encapsulants promote delamination and cracking (Meslier et al., 2024). Inverters can derate or shut down once internal temperatures exceed design thresholds (De Lima et al., 2023), while mounting structures deform under prolonged heat (Okonkwo et al., 2025). Storage systems face reduced efficiency under elevated temperatures (Husain et al., 2024). Beyond heat, PV systems are vulnerable to other extreme events. Hail can cause direct glass and cell damage, with losses up to 21.8% in thin-glass modules (Chakraborty et al., 2023). Strong winds exceeding 200 km/h have detached modules from mounting structures, as observed in northern Italy, Slovenia and Croatia (Bošnjaković et al., 2023). Conversely, moderate winds provide cooling benefits, lowering module temperature by 1–7°C depending on speed (Ataman and Arslanoğlu, 2024). Flooding of ground-mounted systems can erode foundations and damage electrical components, while dust storms increase module temperatures by 5°C and reduce efficiency by 25% after 30 days without cleaning, though up to 90% can be recovered after washing (Chen et al., 2019). Wildfire smoke reduces irradiance significantly, cutting direct normal irradiance (DNI) by up to 42% and global horizontal irradiance (GHI) by up to 17% (Corwin et al., 2025), while increased soiling due to ash accumulation or direct infrastructure damage also affect PV systems (Gilletly et al., 2023; Ford et al., 2024). These stressors also accelerate long-term degradation. Elevated temperatures, UV radiation and arid conditions degrade cables, seals and junction boxes, even when certified for outdoor exposure (IEA, 2022). Prolonged operation above 85°C risks irreversible damage to modules (Gahlot, 2024). In southern Spain and Portugal, degradation rates reach up to 0.8% per year (Ascencio-Vásquez et al., 2019; Poddar et al., 2024). Altogether, these factors compromise both short-term yield and system lifespan, highlighting the need for resilient design and adaptive operation and maintenance (O&M).

Despite the growing risks posed by climate extremes, a substantial gap persists between model-based climate projections and the operational performance of PV systems under stress. The most widely used input for PV yield simulations remains the typical meteorological year (TMY), which condenses long-term averages into a representative year. While useful for estimating typical annual production, TMY files cannot capture short-term anomalies or the dynamics of extreme events such as heatwaves, dust intrusions or wildfire smoke (Loon et al., 2020). As a result, TMY-based simulations often overestimate energy yield and underestimate operational risk. Similarly, many techno-economic models fail to incorporate degradation, soiling or maintenance failures triggered by climatic stressors, further limiting their ability to inform adaptation planning (Hameed et al., 2024). In response, recent literature emphasises the need for a more integrated perspective, bridging climate science, PV modelling and real-world performance analysis. This is reflected in the evolving paradigm of climate resilience, which stresses that solar infrastructure must not only be efficient under average conditions but also capable of withstanding and recovering from disruptive events (IEA, 2022).

We frame PV resilience along three complementary dimensions: resistance, the capacity of components to withstand physical stresses (e.g. hail, wind, heat) without damage (Bošnjaković et al., 2023); robustness, the ability of the plant to sustain acceptable performance under adverse operating conditions (e.g. thermal derating, particulate-laden skies) (Jimba et al., 2025); and recovery, the rapidity and completeness with which performance returns to pre-event levels after disruption (Bošnjaković et al., 2023; Jimba et al., 2025). Building on this framing, three methodological strands are common. Firstly, simulation studies based on TMY remain the baseline and are increasingly complemented with downscaled climate projections from CMIP6/CORDEX to sample future extremes (Rady et al., 2025; Rettie et al., 2023; Gutowski et al., 2020). Secondly, empirical and data-driven analyses couple PV production with observations or reanalyses such as ERA5-Land to evaluate behaviour during real events, using statistical models (Kim et al., 2019; Fregosi and Bolen, 2022), machine learning (Sun et al., 2024; Zhang et al., 2024) and/or digital twins (Pierce et al., 2024). Thirdly, resilience is quantified with dynamic metrics such as resilience curves (Panteli et al., 2017; Dehghani et al., 2023), outage frequency and duration (Ekisheva et al., 2022; Abdelmalak et al., 2023), restoration rates, composite indices (Dwivedi et al., 2025; Ghosh and De, 2022) and survivability metrics estimating the probability of maintaining critical loads (Pizzimbone and Jrad, 2023). While these advances represent important progress, most existing frameworks remain limited in addressing compound or sequential hazards, which are increasingly common under climate change. Bridging this gap requires methodological innovation, region-specific calibration, improved access to high-resolution data and stronger integration of simulations with operational evidence.

As the frequency and intensity of climatic extremes increase, O&M practices become central to the resilience of PV systems. Traditional O&M frameworks, designed for relatively stable conditions, are insufficient when plants are exposed to heatwaves, floods, dust storms, hail and high winds. Preventive strategies, site-specific adaptation and rapid response protocols are essential to safeguard long-term reliability (IEA, 2022). Resilient O&M involves not only regular monitoring and inspections, such as infrared thermography, I–V curve testing and insulation quality assessments, but also the proactive use of early anomaly detection systems. Beyond monitoring, resilience depends on the availability of robust spare-part inventories, adaptive maintenance teams and design provisions that enable rapid recovery after disruptions (IEA, 2022). Adaptation must also be context-specific. In hot and arid regions, failures are commonly caused by high temperatures, strong irradiation and dry conditions; preventive actions include shielding cables from direct exposure, replacing seals regularly and using metal rather than plastic ties. In humid and desert climates, operators face challenges such as potential-induced degradation and accelerated wear due to rapid temperature fluctuations; recommended practices include adjusting inverter settings to local conditions, improving ventilation and frequent filter cleaning (IEA, 2022). In flood-prone areas, elevation of inverters and combiner boxes, IP67-rated materials, reinforced foundations and drainage provisions are key to minimising water ingress and hotspot formation. In cyclone- and storm-prone regions, survival depends on dynamic wind load design, reinforced mounting structures and strict quality control during construction.

Across these contexts, a consistent lesson emerges: resilient PV operation is no longer optional. Extreme weather increasingly shapes the operational reality of European PV plants, requiring O&M strategies to evolve from routine maintenance to climate-adaptive asset management. The challenge is not only to minimise downtime or repair costs, but to ensure that PV generation remains a reliable pillar of Europe’s decarbonisation goals under worsening climate stress.

The aim of this paper is to quantify the impact of heatwaves on the operational performance of PV plants in Southern Europe, focusing on Portugal and Spain. The analysis compares measured performance ratios (PR) before, during and after heatwave events with PVsyst simulations based on two meteorological inputs: the TMY and the ERA5-Land reanalysis data set. The study quantifies heatwave impacts on PV performance in Portugal and Spain by comparing observed PR to PVsyst simulations driven by TMY and ERA5-Land, highlighting event-driven losses, limitations of TMY for design under climate change and the added value of reanalysis. By bridging real-world evidence with simulation outputs, the study contributes to the ongoing discussion on how to integrate climate resilience into PV planning and operation in Europe.

Hourly production data from small/medium-scale PV plants in Southern Europe are used: one in Portugal and two in Spain as shown in Table 1. GHI and air temperature at 2 m time series were extracted from the ERA5-Land (C3S, 2019) reanalysis with hourly resolution and for the grid point closest to the power plant sites. GHI and air temperature for the Amibil site were also available, from a meteorological station of the Agroclimatic Information System for Irrigated Agriculture (SIAR, n.d.). All series are time-zone harmonised (DST-aware). GHI and production data with ERA5-Land GHI < 100W/m² are excluded. The data from 28 July 2024, and 29 July 2024, were excluded in Ariza due to failures of the PV plant, with several periods throughout the day when energy production was 0 kWh.

Heatwave windows were obtained from the official bulletins issued by the Spanish State Meteorological Agency (AEMET) for Spain and the Portuguese Institute for Sea and Atmosphere (IPMA) for Portugal, and two were selected per country for analysis which are shown in Table 2.

In all analyses, we define a pre period as the five days immediately preceding each agency-reported heatwave, and a post-period as the five days immediately following it. For each plant and event, we also compute two within-window descriptors to summarise thermal severity: the maximum hourly 2-m air temperature (T_max) and cumulative degree-hours above 35°C (DH35) as shown in Table 3.

For both Portugal and Spain, the first episode was harsher at every plant, combining higher peaks (T_max up to 38.9°C in Spain, 39.4°C in Portugal) with longer thermal exposure (DH35 up to 39 h in Spain, 26 h in Portugal). Cross-country patterns hint at different stress modes: Spanish events show greater persistence (mean DH35 of 24 h), while Portuguese plants reach higher temperatures (mean T_max of 38.4°C), which may translate into endurance-driven efficiency losses versus peak-temperature risks.

ERA5-Land hourly air temperature and GHI are benchmarked against SIAR for the Amibil location for different regimes (pre-, during and post-heatwave) and for the two Spanish heatwaves, reporting r, mean absolute error (MAE), normalised mean bias error (NMBE) and normalised root mean squared error (NRMSE). Normalised metrics use the station mean as divisor. The goal is to quantify representativeness to support ERA5-Land use.

Each plant is simulated in PVsyst (8.0.14) under two meteorological drivers: a TMY design case using Meteonorm 8.2 (2000–2019), with site-specific horizon (Meteonorm web service) and a 3D scene for shading, and an actual-meteorology case, driven by ERA5-Land GHI and air temperature. Geometry, loss assumptions, shading and the AC boundary are identical across runs. Outputs are hourly simulated energies using the ERA5-Land and TMY data. As defined in IEC 61724-1, the hourly PR is the observed AC energy divided by the simulated AC energy for the same hour (unitless). We also report a daylight-weighted daily PR aggregated from those hours (IEC, 2021).

To characterise typical performance, a diurnal baseline from the five-day pre-heatwave period was computed as the median PR by hour-of-day and a daily baseline as the mean and standard deviation of the daily values for the same period (analogously for TMY). The mean relative difference between TMY and ERA5-Land daily baselines is also computed (Mean $ΔPRrel$⁠). During each event, $W$⁠, hourly anomalies relative to the ERA5-Land diurnal baseline for each site are computed as the difference between the hourly PR during the event and the corresponding baseline value at the same hour and three event metrics are derived: drop magnitude (absolute and relative), which is the most negative PR anomaly during the event; time-to-trough (hours from heatwave start to that minimum); and recovery-to-90% (the first hour after the trough when the observed PR returns to at least 90% of its pre-event baseline at the same hour). These metrics use ERA5-Land in the denominator to keep meteorology explicit and isolate asset behaviour. The same metrics are computed for daily PR values.

The cumulative design-expectation gap is quantified with TMY as energy not served (ENS), which is the positive difference between TMY-simulated energy and observed energy, summed over the event window reported in MWh, MWh/MWp and as a percentage of the TMY-simulated energy over the event.

ERA5-Land hourly air temperature and GHI were benchmarked against a co-located station at the Amibil site, restricting validation to daylight hours (ERA5-Land GHI > 100W/m², Table 3), showing strong agreement. For air temperature, correlations are high in all regimes, bias is essentially neutral throughout (NMBE between −0.16% and 0.39%), and MAE and NRMSE are small with the only notable exceptions being during and post-2S. In practice, a 2–3°C error translates to only around 1% instantaneous efficiency uncertainty for crystalline-Si (using typical 0.4%/°C–0.5%/°C), so temperature representation is adequate for PV performance modelling. For GHI, skill is very high (up to 0.984) and biases are small, but it is event and regime-dependent, with post-2S showing the weakest agreement.

Figure 1 also summarises the point-by-point agreement between ERA5-Land and the co-located station at Amibil across regimes. Air temperature clusters tightly around the 1:1 line, with most points inside the ±10% corridor consistent with the low MAE and NRMSE. GHI shows larger spread matching the larger NRMSE values and the degradation seen post-2S. These results show a minimal temperature bias and small-to-moderate GHI bias, which indicate that ERA5-Land is representative and appropriate as the meteorological driver for the PVsyst runs and as the reference in the PR-based event metrics.

The baseline is the typical diurnal PR for each PV plant, computed from the pre-heatwave window (ERA5-Land driven denominator). A baseline using a TMY-driven denominator is also computed. Figure 2 shows the ERA5-Land baselines by hour of day, while Figure 3 shows the hourly difference between baselines (TMY-ERA5-Land, in PR percentage points). The ERA5-Land diurnal baselines reveal distinct operating signatures. Amibil maintains a high, flat plateau around 95%–98% across the solar peak, Ariza reaches an early maximum and then drifts towards 90%–95% and Zebro operates systematically lower (80%–90%) with a gradual afternoon decline. These shapes frame the later event analysis by fixing the typical hourly performance against which heat-stress anomalies are measured. This is pertinent as Iberian hot extremes are increasing under anthropogenic warming (Vautard et al., 2023; EEA, 2025).

Figure 3 indicates that ERA5-Land and TMY baselines are broadly aligned at mid-day, with the highest differences concentrated near sunrise and sunset. Because these hours yield less energy, hourly differences translate only weakly to the daily-energy weighted PR.

Table 4 reports daily, energy-weighted PR (mean ± standard deviation, SD) for ERA5-Land and TMY together with the relative mean difference. The ERA5-Land daily baselines are internally consistent and narrow. Amibil centres near 95% with small dispersion, Ariza is slightly lower but also has small dispersion and Zebro operates at lower levels (83%–86%) with broader dispersion reflecting its lower, more sloped diurnal shape. The hourly profile of the baselines (Figure 2) suggests that offsets arise mainly at hours closer to sunrise and sunset, hence their impact on daily, energy-weighted PR is limited but not negligible when deviations are large (2S at Amibil and Zebro). Values above 100% arise because this index is defined as observed energy divided by the TMY simulation for the same period. They simply indicate that the TMY driver underestimates energy in the pre-window (mainly at low sun angles) and do not imply physical PR above 100%.

Performance during the events is assessed by hourly and daily PR anomalies relative to the ERA5-Land based diurnal baseline from the pre-event window. Figures 4 through 6 show these anomalies as well as the trough (largest negative anomaly) and the recovery-to-90%. In Table 5, resilience metrics are condensed for hourly and daily data. Large positive spikes at sunrise/sunset reflect the relative anomaly definition at low-sun hours (small hourly baseline/simulated energy), not physical PR > 100%. These points have negligible energy weight and do not affect daily metrics.

Analysing Figure 4 and Table 3 for Amibil, both events exhibit a classic heat-stress signature: troughs occur in the afternoon when module/inverter temperatures peak, followed by rapid recovery as thermal load eases. The deeper but shorter-lived shock in 1S is consistent with its higher thermal severity (Table 6), likely reflecting brief protective derating near the peak and quick restoration as temperatures fall. In 2S, the trough occurs at 17:00 UTC with 1 h of daylight remaining; because our evaluation uses daylight-only hours, the reported recovery-to-90% includes the overnight gap and should not be interpreted as a slower physical response. This pattern is consistent with temperature-driven efficiency loss and inverter thermal derating near peak component temperatures (Ebhota and Tabakov, 2022; De Lima et al., 2023). Amibil’s flat, high midday baseline means sunrise/sunset anomalies carry reduced energy weight, so large hourly excursions compress into small daily deficits. At the daily scale, both events show single-digit drops that peak on Day 4 (1S) and Day 2 (2S) and recover within the same day, indicating that heat-related risk is concentrated in a handful of hot afternoon hours rather than sustained multi-day degradation.

Ariza’s response mirrors its baseline signature as shown in Figure 5 (early peak followed by an afternoon drift), with PR anomalies clustering near zero for much of the window but having extreme values in mid-afternoon, when thermal load and inverter temperatures are highest. In 1S, the drop is moderate and short-lived, with an almost immediate rebound once the peak passes, which is consistent with brief protective derating rather than sustained loss. By contrast, 2S shows a deeper trough and a slower intra-day recovery, even though its thermal severity is lower (Table 6). This inversion suggests that factors such as inverter derating thresholds, transient curtailment or weaker convective cooling, dominated over meteorology in determining the depth and duration of the dip. Such plant-specific sensitivity accords with PV vulnerability literature and O&M guidance (Patt et al., 2013; IEA, 2022). At the daily scale, losses remain single-digit and short-lived (Table 5), reinforcing that risk at Ariza is hour-centric, concentrated in hot afternoon hours rather than a multi-day decline.

Zebro behaves in line with its sloping afternoon baseline, as seen in Figure 6. In 1P, Zebro’s hourly anomaly has a convex diurnal shape, slightly positive at midday and negative towards sunrise/sunset, while the hourly trough occurs late afternoon and resolves the next day. Because the negative hours coincide with low irradiance, the daily aggregate does not turn negative during the heatwave period (^†in Table 5 marks the daily minimum just after the heatwave). In 2P, however, the trough occurs near solar noon, carries high energy weight and translates into a daily drop on Day 1 with recovery within that day. Thus, at Zebro the deficit is not systematically midday centred.

Across the three plants the response is hour-centric (sharp troughs with intra-day recovery), but the magnitude and day-level impact depend on each site’s baseline and on when within the diurnal cycle the minimum occurs. The intra-day drops and recoveries observed even in the post-heatwave regimes indicate that the proximate driver of hourly PR deficits is not the label “heatwave” per se, but short-lived exceedance of thermal thresholds (high air temperature with limited wind cooling) that triggers protective derating and then normalisation once thermal load eases. As Western European heat extremes intensify, such hot-hour risks will occur more often in Iberia (Vautard et al., 2023; Dong and Sutton, 2025; EEA, 2025). This is consistent with PV temperature physics, where module temperature responds to irradiance, ambient temperature and wind speed, and with evidence that inverters thermally derate under high component/ambient temperatures; hence, acute afternoon spikes can produce the same signature inside or outside officially declared heatwaves.

Table 7 quantifies the ENS when TMY is used as the driver, i.e. the energy that TMY-based design would expect but was not delivered in operation. In all cases these values are non-zero, but their magnitude and timing vary widely across plants. In absolute terms and per-kWp, the largest gaps occur at Zebro-2P, Amibil-1S and Amibil-2S. As a share of simulated energy, the largest relative shortfall is Ariza-2S, followed by Zebro-2P and Amibil-2S. The regime split is also site and heatwave dependent. During-event losses dominate at Amibil-1S, Zebro-1P and Zebro-2P, while post-event losses are the largest component for Amibil-2S and Ariza-2S. This indicates that TMY-only simulations can overestimate both during and, in some cases, after the event. This aligns with evidence that TMY under-represents extremes and so event-centred stress tests are warranted (Loon et al., 2020; Perera et al., 2020). Within this limited sample (three plants, four events), the data show that the size and timing of the TMY gap are plant-specific and event-specific, reinforcing the need to complement TMY with event-centred analyses when assessing operational risk. Future hot-hour frequencies can be quantified using high-resolution climate projections (Gutowski et al., 2020).

Heatwaves, projected to intensify across Iberia, pose operational risks for PV plants. We assessed three Iberian plants across four agency-declared events by pairing observed energy with ERA5-Land-driven and TMY PVsyst simulations in an event-centred framework. ERA5-Land reproduced site meteorology well at the validated site (air temperature r = 0.91–0.95; GHI up to 0.98), supporting its use as the denominator for PR metrics. Pre-event PR baselines revealed distinct operating signatures, which shaped event responses by fixing the typical hourly performance against which heat-stress anomalies were measured. Across events, impacts were hour-centric: sharp afternoon PR troughs with intra-day recovery. The deepest hourly loss reached 90.44% (Amibil-1S), and the largest daily loss was 17.59% (Zebro-2P) with recovery within 0–1 days. TMY expectations overestimated energy in all cases, with the largest absolute gap at Zebro-2P and the highest relative gap at Ariza-2S. In planning and contracting terms, this means design-year expectations systematically understate heat-related operational risk. Intra-day drops and recoveries also occurred outside official heatwave windows, indicating that brief exceedances of thermal thresholds, not the label “heatwave” per se, drive loss concentration in hot hours. These findings are bounded by scope (three plants, four events, in situ validation at one site) and by comparing a stationary TMY (2000–2019) with 2024–2025 conditions. Even so, the message for European deployment is practical and climate-relevant. Complement TMY with event-centred stress tests using reanalysis or site-calibrated data, specify and operate for hot-hour resilience (thermal headroom, ventilation, derating management, rapid post-event inspection) and recognise in adequacy and contracts that short, frequent hot-hour losses will occur more often as the climate warms.

The metrics developed here provide actionable triggers for today’s decisions. Developers can use ENS and trough timing to set minimum thermal headroom and ventilation/layout requirements in procurement. Operators can codify hot-hour protocols (derating thresholds, rapid post-event inspection) and report ENS alongside PR during extreme heat to make losses visible in dashboards. Planners and regulators can account for more frequent hot-hour deficits in adequacy and reserve margins, recognising that typical-year expectations systematically understate risk as the climate warms. These actions are implementable with existing SCADA data and reanalysis forcing. Scaling the framework across plant typologies and regions and coupling it with high-resolution climate projections will quantify how often high-risk hot hours are likely to occur, enabling measurable resilience gains in procurement, design reviews and operational dashboards.