PowerVeritas
Insights / Field note · Solar

Solar panels like the sun but not the heat

Sunny days bring more light, but they also bring hotter panels. Three months of five-minute data from a solar site show why the second effect is easy to misread, and what it costs across a year.

← All insights

It is obvious that a solar panel produces more power in bright sun. Less obvious is that the heat which comes with it works the other way. Panels lose efficiency as they warm, and in strong sun they run well above the temperature of the air around them.

To see the effect we plotted three months of five-minute data from a solar site we monitor against satellite irradiance, with every point coloured by air temperature.

Scatter plot of solar site AC power against horizontal irradiance, points coloured by air temperature. The data splits into two arms, a cooler upper arm labelled morning and a warmer lower arm labelled afternoon, with a flat band of points at the top labelled as the inverter output limit.
Three months of five-minute output against horizontal irradiance, coloured by air temperature. The scatter splits into two arms, with a flat band across the top where the inverter is limiting its own output.

The first plot is not the temperature story

The data separates into two clear arms. Cooler morning points ride high, warmer afternoon points sit low, which looks exactly like the temperature effect. The real cause is the angle of the sun. This array faces south-east, so morning sun strikes it closer to square-on. Measured against horizontal irradiance, the same sunlight produces around 10 kW more in the morning than the afternoon. Running the site physics model with everything held fixed except the time of day reproduces that gap from geometry alone. Temperature accounts for under 1 kW of the roughly 10 kW separation.

Correct for the sun’s direction first

The fix is to convert horizontal irradiance into the energy actually arriving on the panel face, the plane-of-array value, using the sun’s position with the array’s tilt and orientation. On the corrected axis the two arms collapse onto a single band, agreeing to within about 1 kW at matched irradiance. What is left is temperature.

The same scatter plot with irradiance corrected to the plane of the panels. The two arms have collapsed onto a single band, with warm orange points sitting below cool teal points, between two dashed lines showing expected output at 10 and 30 degrees Celsius air temperature.
The same site with irradiance corrected to the panel face, clear sky days only. Dashed lines are the expected output at 10 °C and 30 °C air.

What the data shows

  • Warm points sit below cool points at the same panel-plane irradiance, tracking the modelled physics lines.
  • A conventional crystalline-silicon module loses around 0.3% to 0.4% of its rated output for every degree its cells rise above the 25 °C rating condition, and in strong sun the cells here run 15 to 25 °C above the air, depending mostly on wind.
  • Reading the two modelled lines, output at the same panel-plane irradiance is about 8% lower on a 30 °C day than a 10 °C day. The measured points show a slightly larger gap.
  • The loss is temporary and fully reversible. Panels recover as soon as they cool, this is not degradation.
Method notes. Plant output is recorded at five-minute intervals. Irradiance comes from Open-Meteo’s satellite radiation product at its native resolution, around ten minutes for Europe, and air temperature and wind from the hourly reanalysis archive. Both are time-aligned to the plant records by interpolation. Neither is measured on site, so individual short-interval values carry appreciable uncertainty. The charts show clear sky days only, with unsteady samples excluded. Records affected by an inverter output limit are kept in the first chart, where they are labelled, and excluded from the charts that follow it. Irradiance is corrected from the horizontal onto the plane of the panels using the sun’s position and the array geometry, and cell temperature is modelled from irradiance, air temperature and wind using established industry methods.

What it costs across a year

Applying the same physics to typical local weather, month by month, gives the seasonal picture.

Line chart of modelled temperature efficiency by month, peaking around 5 percent above the 25 degree rating in December and January and falling to 5.5 percent below in July, with sun-hour weighted cell temperature on a second axis mirroring the shape.
Modelled clear-day temperature efficiency through the year for this site, against the panel’s 25 °C rating, with sun-hour weighted cell temperature alongside.

On temperature alone, a UK array spends roughly half the year with its cells below the 25 °C rating condition, so the temperature correction is positive through the colder months and negative in summer. For this site the modelled factor is about 5% positive in December and January and about 5.5% negative in July, a seasonal swing of roughly 10 percentage points. Actual output still depends on irradiance, incidence angle, system losses and the inverter’s own limits. It is why a plant judged against a flat rating can look like it is underperforming every summer and over-performing every winter, when both are simply expected behaviour.

Modelling the site, five minutes at a time

PowerVeritas built its own monitoring and modelling software for exactly this problem. It pulls plant telemetry and weather data automatically, then builds a physics model of the specific site rather than a generic one. Array tilt, orientation and system losses are fitted against the plant’s own production history and constrained by what is known about the installation, so the model reflects how the site was actually built rather than what the design drawing assumed. Those parameters can trade off against one another in a fit, so the result is checked against site records rather than taken on the numbers alone.

For every five-minute record the model works through the chain used in this article. The sun’s position, the irradiance arriving on the panel face, the cell temperature from irradiance with air temperature and wind, the resulting efficiency loss, and finally the inverter’s own limits. The result is an expected output for that moment, plotted alongside what the plant actually did.

Time series for one day showing measured AC power and the model prediction tracking each other closely from dawn to dusk, with a shaded band through the middle of the day where the inverter limited its output below the prediction, and a mode strip beneath confirming the throttled period.
One day of measured output against the model’s prediction in the PowerVeritas monitoring system. The two track each other from first light to dusk. The shaded band from mid-morning to early afternoon is the inverter limiting its own output, where the prediction shows what the array could otherwise have produced. A small number of gap-filled records are marked.

With that in place the number worth looking at is the difference between expected and actual. Weather, sun angle and heat are already accounted for, so the residual can be investigated for soiling, shading, faults, curtailment or genuine degradation, while allowing for telemetry and model uncertainty. Those are the things worth acting on.

And then there is the inverter

The shaded band in that chart is the second half of the story. Inverters lose efficiency as they warm, and above a threshold they deliberately reduce output to protect themselves. That is a power cap rather than an efficiency loss, and it appears as the flat ceiling labelled in the first chart on this page, well below what the conditions allowed. Those records are excluded from the charts that follow it, so they do not distort the temperature picture.

A flat ceiling on its own does not prove thermal derating. It can equally be a configured export limit or the inverter simply clipping at its rating, so the operating mode and temperature telemetry are what separate them. Where those confirm thermal derating, repeated events can point to blocked filters, a failed fan or excessive enclosure temperature, and the lost production is often recoverable.

Separating expected environmental behaviour from genuine underperformance is the foundation of any performance assessment. Without that correction a hot summer reads as a fault and a cold winter reads as a win, when neither is true.

Site telemetry used with the owner’s permission. Irradiance and weather from the Open-Meteo archive. Analysis and charts by PowerVeritas.