Solar panel degradation, and what your warranty covers

Where "0.5% a year" comes from, and what it actually says

Nearly every claim about panel degradation traces to one NREL paper, Photovoltaic Degradation Rates: An Analytical Review. It is careful work and it is quoted in a way it does not support.

First: it is a literature review, not a measurement campaign. It assembles degradation rates reported in published field-test literature "throughout the last 40 years", so roughly 1972 to 2012. Each of its 1,920 entries is one published measurement of one module or one system, not one panel-year of observation.

Second: 0.5% is the median, and the mean is 0.8%. The paper says why, verbatim: "The distribution is skewed toward high degradation rates with a mean of 0.8%/year and a median of 0.5%/year." Quoting 0.5% as "the average" is simply wrong. The average is 60% higher than the number everyone repeats.

Third: the spread is wide and the paper gives one figure for it. 78% of all reported rates are below 1% a year. By implication, more than one in five published rates is at or above 1%. There is no standard deviation published for the pooled distribution, so that 78% is the only dispersion figure anyone can honestly cite.

Fourth: it is a crystalline-silicon number. 1,751 of the 1,920 rates are crystalline silicon, which is 91% of the sample. Thin film contributes 169 rates with a median of 1.0% and a mean of 1.5%. The paper's own conclusion notes thin film rates are "statistically closer to 1%/year than to the 0·5%/year necessary to meet the 25-year commercial warranties."

The warranty is a floor, not a forecast

Performance warranties are usually presented as though they describe expected behaviour. They do not. They describe the line below which the manufacturer accepts a claims risk, and that line is set deliberately below where most modules are expected to sit.

The arithmetic makes it obvious. One manufacturer's current warranty caps degradation at 0.25% a year after year one. The median field rate in the NREL review is 0.5%, and the mean is 0.8%. The warranty cap is half the median and roughly a third of the mean.

That is not a contradiction and it is not a manufacturer being generous. Modern modules genuinely perform better than a sample dominated by pre-2000 hardware. But it does mean the warranted line and the expected line are different objects, and a proposal that presents the warranty curve as a production forecast is showing you the wrong one.

The baselines also differ between manufacturers, and this is easy to miss. One warrants a percentage of the module's nameplate power. Another warrants a percentage of the minimum power specified in the data sheet. On a module sold with a positive-only tolerance those coincide. On a module with a plus-or-minus tolerance they do not, and the second baseline is the lower one. Two warranties quoting the same percentage can therefore promise different amounts of power.

The measurement tolerance is larger than a decade of the guarantee

The measurement tolerance is the finding that changes how the whole warranty document should be read, and it is not hidden. It is in the same sentence as the promise.

The REC Alpha warranty provides that its remedies apply if the product "does not reach the warranted power output levels set out above when measured by the Warrantor or by an accredited independent measuring institute agreed to prior to testing by the Warrantor, under standard test conditions (IEC 61215) and taking into account a ±3% tolerance range".

Now put that beside the guarantee in the same document: degradation of no more than 0.25% a year.

Three percent divided by 0.25% a year is twelve years. A module can sit measurably below its warranted line for more than a decade and still fall inside the stated tolerance band of the test that determines whether it is below the line.

We want to be precise about what that is and is not. It is arithmetic on two figures in the same clause of a published warranty. It is not an allegation about how any manufacturer behaves, we have no evidence about claims handling, and a tolerance band exists for a legitimate reason: measuring module power is genuinely difficult and no laboratory returns an exact figure. But a buyer comparing "0.25% a year" against "0.54% a year" as though the difference were the whole story is comparing guarantees whose enforcement threshold is dominated by a term neither number contains.

And there is a second measurement finding, established by absence. None of the four manufacturer documents we have now read references IEC 61853, the standard for energy rating across multiple irradiance and temperature conditions. Every performance warranty we read is written against a single standard-test-conditions point. Your roof is not a single point, and the standard that describes performance across real conditions is not the one your warranty is enforced under.

What a successful claim actually gets you

Suppose you clear all of that: the shortfall is real, it is outside the tolerance, and the manufacturer agrees. Three features of the remedy are worth knowing before you rely on it.

The manufacturer chooses the remedy, not you. The options are typically to repair, to replace with an equivalent product, to supply additional panels sufficient to reach the warranted output, or to refund. All of them are expressed as being at the manufacturer's sole option.

A cash remedy is benchmarked to today's prices, not yours. REC's wording is to "refund the current market price of an equivalent product at the time of the claim". Module prices have fallen substantially over the period these warranties run. A cash settlement in year eighteen is priced against an eighteen-years-cheaper market, not against what you paid.

Replacements can be refurbished, and the clock does not restart. The warrantor "may use remanufactured or refurbished parts or products", and the warranty periods "will not be extended in any way in the event of a replacement or repair". A module replaced in year twenty carries five years of cover, not twenty-five.

And labour is yours. Every manufacturer document we have read across this cluster, now four of them, excludes the cost of removing and reinstalling. The one exception we found is an optional extended package from one manufacturer that adds labour cover, and it is contingent on the installer registering the system, which means it depends on a third party doing something correctly at the time of installation.