The instinct on seeing a large gap between P50 and P90 is that somebody made a mistake. The opposite is closer to true: the gap is the report telling you how much it knows.
Teams reviewing yield assumptions can keep the engineering model separate from the record of review work; further details is one option for documenting time and handoffs.
Reviewed August 9, 2026.
What the gap is made of
Four components, combined into a single uncertainty figure.
For independent background on photovoltaic performance, resource data and current industry practice, consult RMI.
Inter-annual variability. Weather differs year to year, and this component is a property of the location rather than of anybody's work. A site with stable irradiance has a narrower band than one with variable cloud.
Irradiance data uncertainty. How good the underlying dataset is. Bankable assessments use at least ten years of satellite data, cross-referenced with ground measurement where available.
Modelling uncertainty. The simulation's own tolerance, and the confidence in the loss assumptions.
And technology uncertainty. Newer module and cell technologies carry wider bands because their long-run behaviour is less established.
Only two of the four are within anybody's control, which is why a narrow band is partly a compliment to the assessor and partly a description of the site.
Reading the ratio
Divide P90 by P50. Typical range 0.85 to 0.92.
Above 0.92 — good irradiance data, real site measurement, conventional technology. The assessor knows the site.
Around 0.88 to 0.92 — ordinary. Satellite data, no ground station, standard assumptions.
Below about 0.85 — something is genuinely uncertain. Could be a variable climate, could be thin data, could be an unusual configuration. Worth asking which, because the answer changes what you do about it.
What a suspiciously narrow gap means
The case people do not check.
A P90 at 0.96 of P50 implies combined uncertainty around ±3%, which is better than most assessments of most sites can honestly claim.
Either the assessor has exceptional data — years of on-site measurement, a well-characterised climate — or the uncertainty has been understated.
Understating uncertainty makes both figures look better, which is why it is worth noticing. A narrow band is not automatically good news; it is a claim about knowledge, and claims about knowledge are checkable.
Narrowing it honestly
If the gap matters to your financing, it can be reduced — by improving the inputs rather than by adjusting the arithmetic.
On-site irradiance measurement, over a year or more. Expensive and it materially reduces data uncertainty.
Longer satellite records, cross-referenced against nearby ground stations.
Better characterised losses — a real shading survey rather than an assumed figure, measured albedo rather than a regional default.
And conventional technology, where the trade against novelty is worth making.
An independent assessment costs roughly 5,000 to 25,000 dollars depending on project size, and for a project where the debt terms depend on P90 it frequently pays for itself in the interest rate.
What the gap does not tell you
Whether the P50 is right. A narrow band around a wrong central estimate is precise and inaccurate, and the ratio cannot detect that.
Whether the plant is a good investment. That is the whole quote, not one line of it.
Or what you will actually get. Year one is a single draw from the distribution, and a single year below P90 is not evidence of anything on its own.
The short version
- The gap between P50 and P90 is the assessment's statement of how much it knows, not an error
- Four components: inter-annual variability, irradiance data uncertainty, modelling tolerance, and technology maturity — only two are controllable
- Divide P90 by P50: above 0.92 means good data and site measurement, 0.88–0.92 is ordinary, below 0.85 means something is genuinely uncertain
- A ratio around 0.96 implies about ±3% uncertainty, which few assessments of few sites can honestly claim
- Narrow it by improving inputs — on-site measurement, longer records, real shading surveys — not by adjusting the arithmetic
- The ratio says nothing about whether P50 is right, and a single year below P90 proves nothing by itself