What P50 and P90 mean
P50 and P90 are probability-of-exceedance estimates for a specified solar-energy output and period. P50 has a 50% chance of being exceeded under the stated model; P90 has a 90% chance. P90 is therefore normally lower than P50. Neither is a promised minimum for the next operating year.
State the period and the meter boundary. A one-year P90 for energy at the AC meter is not the same as a lifetime P90, a DC-array forecast or a revenue guarantee.
The same idea extends to other probabilities. P75 is exceeded with 75% probability, P95 with 95% and P99 with 99%, so each is a more conservative figure than the one before. P50 is the median of the modelled distribution, not a best case. P90 is cautious, not a worst case, because one outcome in ten is expected to fall below it.
Why a single yield figure is not enough
A solar plant does not produce the same energy every year. Weather varies, the irradiation dataset behind the forecast has its own error, the model makes assumptions about losses, and the plant ages. A forecast that gives one number hides all of this. Stating a P50 together with a P90 tells the reader how far the result could fall short and how likely that is, which is the information a lender, insurer or owner needs to judge risk. The solar yield guide covers the underlying energy estimate.
How the estimate is built
- Collect the solar-resource history for the site, ideally a multi-year series from a validated source, and note its stated uncertainty.
- Run a yield simulation with the proposed layout (check the strings first with the string sizing calculator), tilt, shading, equipment and loss assumptions to get the central (P50) annual energy. The yield simulation software guide explains how such models work.
- List the uncertainty components, each expressed as a relative standard deviation: year-to-year resource variability, resource-data uncertainty, model and loss uncertainty, and degradation or availability uncertainty where relevant.
- Combine independent components into one value, usually by root-sum-of-squares.
- Read the percentiles from the resulting distribution.
An illustrative calculation
If the distribution is assumed to be normal, a P-value for exceedance probability p is:
where σ is the combined relative standard deviation and z is the standard normal value for that probability: about 0.674 for P75, 1.282 for P90, 1.645 for P95 and 2.326 for P99.
| Value | Exceeded with probability | z | Annual energy (MWh) |
|---|---|---|---|
| P50 | 50% | 0 | 1,000 |
| P75 | 75% | 0.674 | 964 |
| P90 | 90% | 1.282 | 931 |
| P95 | 95% | 1.645 | 911 |
| P99 | 99% | 2.326 | 874 |
How the period changes the answer
The weather component shrinks when you average over more years, because good and poor years offset one another. A P90 for the ten-year average is therefore closer to P50 than a P90 for any single year. Model and dataset uncertainty do not shrink in the same way, since they are systematic. A report should say whether its P90 is a one-year or a ten-year figure, and a buyer comparing two reports should check that both use the same basis.
What should a buyer review?
| Input or assumption | Why it changes the estimate |
|---|---|
| Solar-resource history and location | Affects central estimate and weather uncertainty |
| Plane-of-array and shading model | Converts mapped resource to the proposed geometry |
| Soiling, temperature, electrical and availability losses | Determines delivered energy rather than theoretical module output |
| Degradation and study period | Alters long-term energy and uncertainty |
| Correlations and uncertainty method | Determines the P50-to-P90 spread |
A lender may commission an independent energy assessment and may size debt conservatively, but the exact required percentile is a transaction decision. Do not claim that every lender mandates a particular software tool or data-history length without that lender’s specification. The irradiance data guide explains resource inputs; CUF is a separate way to express energy relative to capacity.
Practical request: ask for the model version, weather dataset, loss table, uncertainty report, energy boundary and reviewer identity along with any P50/P90 figures.
Who uses which value
| Use | Value commonly looked at |
|---|---|
| Owner planning and customer-facing energy forecast | P50, with the assumptions stated |
| Debt sizing and lender stress testing | P90 or a more conservative value, set per transaction |
| Sensitivity checks and insurance discussions | P90, P95 or P99 depending on the counterparty |
This table describes common practice, not a rule. Check the term sheet, the independent engineer’s scope and the offtake contract for the value that actually applies to a given project.
P50, the average and the metered result
P50 is the median, while the average (mean) is a different statistic. For a symmetric distribution they are the same. In practice the useful check is the metered output after commissioning. Compare it with the forecast on the same basis, correcting for the weather that actually occurred, using the performance-monitoring approach in IEC 61724-1. A single year below the P50 forecast is not evidence of a bad forecast, and a single year above P90 is not evidence of a good one. A few years of operating data tell you more.
How P-values relate to CUF and performance ratio
Capacity utilisation factor expresses annual energy as a share of what the plant would produce at nameplate output all year. Performance ratio compares actual output with the output the measured irradiation should deliver. Both describe an energy result. P50 and P90 attach a probability to the forecast of that result. A report that quotes a CUF without saying whether it is P50, P90 or a measured value is incomplete.
Where this applies
Project finance and debt sizing for utility and large captive plants, offtake and tariff modelling, equity return analysis, insurance structuring and independent engineer reviews. For a home or small commercial rooftop, a transparent annual estimate with its assumptions and a comparison against metered generation is usually more useful than a formal probability analysis.
Common mistakes
- Quoting a P50 and a P90 without saying whether they are one-year or multi-year.
- Treating a P90 as a guaranteed minimum.
- Copying a P90-to-P50 ratio from another site or another report.
- Comparing P-values from different datasets without checking the dataset uncertainty.
- Using a P-value from DC array output where the contract is measured at the AC meter.
- Ignoring degradation or availability assumptions when comparing lifetime figures.
Standards and references
IEC 61724-1 covers PV system performance monitoring and is the usual basis for comparing measured results with a forecast. Methods for energy yield assessment and uncertainty are set out by independent engineers and the yield-modelling tools they use, not by a single national standard, so the report itself must state its method.
Key takeaways
P50 and P90 are probability statements about a stated energy forecast over a stated period. P90 is not a fixed percentage of P50, and it is not a guarantee. To use one, you need the period, the meter boundary, the dataset, the loss and uncertainty assumptions, and the reviewer’s identity.
Frequently Asked Questions
What is P50 solar yield?
What is P90 solar yield?
Why are P-values used instead of one yield figure?
Is P90 always a fixed percentage of P50?
How is P90 calculated?
Are P50 and P90 best-case and worst-case?
What is P75, P95 or P99?
Who uses P50 and who uses P90?
Is P90 the same for one year and for ten years?
Is P50 the same as the average yield?
What inputs move the P50 to P90 gap?
Is a P90 report automatically bankable?
How are P50 and P90 related to CUF and performance ratio?
Do P-values apply to a home rooftop system?
What should I ask for along with a P50/P90 figure?
From the Blog
Co-founder & CEO driving Qbits Energy's product strategy and growth across India.