P50/P90 Energy Yield Explained: How to Interpret Solar Production Estimates

A 5 MWp solar plant modeled with a P50 yield of 8,400 MWh per year and a P90 of 7,700 MWh produces, on average, 700 MWh of difference between the central forecast and the bankable downside. That gap is not error. It is the visible fingerprint of weather variability, dataset uncertainty, and simulation tolerance, expressed as a probability. Misreading it is the single most common reason that solar projects miss their first-year production targets and trigger debt-service covenant warnings.

P50 and P90 are the two numbers that determine whether a solar project gets financed, how much debt it can carry, and what the developer is willing to guarantee under a power purchase agreement. Yet in our experience, fewer than one in three sales engineers and project developers can explain the difference correctly when asked. This guide closes that gap. It walks through what P50 and P90 mean as exceedance probabilities, how they are calculated from irradiance data and uncertainty inputs, and how to read a yield report without falling into the traps that derail proposals at the lender stage.

In this guide:

• What P50, P75, P90, P95, and P99 mean as exceedance probabilities
• The math behind P-value calculations, including the standard deviation method
• How weather data, simulation models, and inter-annual variability stack into total uncertainty
• A worked P90 example for a 5 MWp project with realistic numbers
• How banks, equity investors, and PPA counterparties interpret these values
• How to read a yield assessment report without the common misreadings
• How solar design software like PVsyst, Solargis, and SurgePV generates P-values
• Action items for designers, EPCs, and finance teams

What P50 and P90 Mean: Exceedance Probabilities Defined

P50 and P90 are exceedance probabilities. They are not best-case and worst-case scenarios. They are not optimistic and pessimistic forecasts. They are statistical statements about how likely a given annual production number is to be met or beaten in any single year of operation.

The plain-English definition for each value:

• P50 — Annual energy yield that has a 50% chance of being exceeded in any given year. Half of years produce more, half produce less.
• P75 — Annual yield exceeded in 3 out of 4 years (75% of the time).
• P90 — Annual yield exceeded in 9 out of 10 years.
• P95 — Annual yield exceeded in 19 out of 20 years.
• P99 — Annual yield exceeded in 99 out of 100 years.

The “P” stands for probability of exceedance. A higher P-number means a more conservative (lower) energy estimate, because there is a higher probability of clearing it. P90 is more conservative than P50 because it is harder to fall below a value that 90% of years exceed.

The numbers come from a Gaussian (normal) distribution fitted to the expected annual yields. P50 is the mean of that distribution. P90, P95, and P99 sit progressively further down the left tail.

Why The Naming Trips People Up

The first instinct for many people new to solar finance is to read “P90” as “90% of expected production.” That is wrong. P90 is the production level that the system is 90% likely to achieve or exceed, not 90% of P50.

The actual relationship depends on the project’s combined uncertainty. A project with ±6% total uncertainty produces a P90 about 7.7% below P50. A project with ±10% total uncertainty produces a P90 about 12.8% below P50. The gap widens as uncertainty rises.

How P50, P75, P90, P95, and P99 Differ in Practice

Each P-value corresponds to a specific shift, in standard deviations, below the mean of the yield distribution. The shifts come from the inverse normal cumulative distribution function. The five values used in solar finance are:

The formula is the same for every P-value:

P_xx = P50 − (z-score × σ_total)

Where σ_total is the total uncertainty in kWh (or as a fraction of P50). This is the single equation that drives every probabilistic yield report on the market. Once you have P50 and total uncertainty, every other P-value is a direct calculation.

A Worked Example for All Five Values

Take a project with a P50 of 10,000 MWh/year and total combined uncertainty of 7%, which means σ_total = 700 MWh. The full P-curve looks like this:

Each row is the same formula with a different z-score. P75 sits 472 MWh below P50. P99 sits 1,628 MWh below P50. The spread between P50 and P99 is a direct readout of how confident the model is in its central forecast.

This is why two projects with identical P50 numbers can have very different P90s. A well-instrumented site with 25 years of ground-station data and a high-confidence simulation might produce a P50 of 10,000 MWh and a P90 of 9,300 MWh (93% of P50). A new-build site relying on satellite-only data with no on-site measurements could produce the same P50 of 10,000 MWh but a P90 of only 8,800 MWh (88% of P50). The lender will treat those two projects very differently, even though their headline numbers match.

How P-Values Are Calculated: From Irradiance Data to Probability

A bankable P50/P90 calculation is the output of a five-step workflow that turns raw irradiance data into a probability distribution. Every commercial yield report follows the same skeleton. Understanding it makes every report easier to read, audit, and compare.

Step 1 — Gather Long-Term Irradiance Data

The starting point is multi-year solar resource data for the project location. Industry standard is 15 to 25 years of hourly irradiance, broken into:

• GHI (Global Horizontal Irradiance) — total solar energy hitting a horizontal surface
• DNI (Direct Normal Irradiance) — beam radiation that arrives in a straight line from the sun
• DHI (Diffuse Horizontal Irradiance) — radiation scattered by the atmosphere

Datasets come from satellite providers like Solargis, Meteonorm, NSRDB, SolarAnywhere, and ground measurement networks. For a deeper walk-through of the three irradiance components, see our solar irradiance: GHI, DNI, DHI guide.

The irradiance dataset feeds two parallel calculations:

1. Time series simulation — every hour of every year of weather data is run through the PV system model to produce a year-by-year energy yield series. The mean of that series is P50. The standard deviation of that series captures inter-annual variability.
2. Typical Meteorological Year (TMY) — a synthetic 8,760-hour year built from “typical” months across the historical period. TMY runs are faster but underestimate variability if used alone.

Most modern bankable reports run a full time-series simulation. TMY-based shortcuts can produce a P90 estimate that is 4% off the time-series benchmark, which is enough to flip a debt-coverage calculation.

Step 2 — Run the System Simulation

The PV model takes the irradiance data plus the system definition (panel type, tilt, azimuth, inverter characteristics, soiling, snow, shading, wiring losses, transformer losses) and produces a year-by-year AC energy output series.

Loss components that materially shift P50 include:

• Module degradation — typically 0.4 to 0.7% per year for monofacial silicon, 0.3 to 0.5% for high-quality bifacial. See our entry on annual degradation rate for technology-specific numbers.
• DC and AC losses — wiring, mismatch, transformer, auxiliary loads — often 4 to 7% combined
• Soiling — 1 to 5% depending on local dust, pollen, and rainfall patterns
• Shading and inter-row losses — minimized through careful design and validated with shadow analysis
• Inverter clipping — energy lost when DC output exceeds AC inverter capacity

Each loss line is either deterministic (fixed assumption) or stochastic (carries an uncertainty band that flows into σ_total).

Step 3 — Quantify Each Uncertainty Source

This is the analytical heart of a bankable yield report. Total uncertainty is the root-sum-square of every independent uncertainty contribution. The standard categories are:

Each of these is independent and gets squared and summed:

σ_total = √(σ_irradiance² + σ_model² + σ_interannual² + σ_local² + σ_module²)

For a typical commercial-scale European PV project, total combined uncertainty lands between 6 and 9% of P50. For an instrumented utility-scale project with multi-year on-site pyranometer data, it can drop to 4.5 to 6%. For a residential system in a high-variability climate with no on-site measurements, it can rise to 10 to 12%.

Step 4 — Build the Gaussian Distribution

With P50 (mean) and σ_total (combined standard deviation), the report builds the normal distribution function. Excel’s NORM.INV(probability, P50, σ_total) returns any P-value directly. PVsyst, Solargis, RatedPower, HelioScope, and SurgePV automate this step.

Step 5 — Output the P-Value Table

The final report shows P50 alongside P75, P90, P95, and P99 — the values lenders, equity investors, and PPA counterparties care about. Some reports also show a 1-year P90 and a 10-year P90, where the 10-year value benefits from inter-annual smoothing.

Sources of Uncertainty in a Yield Estimate

The P50/P90 gap is a direct readout of total project uncertainty. Knowing where uncertainty comes from is what separates sales engineers who can defend a yield number to a lender from those who cannot.

Weather Data Uncertainty

The dominant uncertainty source for most projects is the long-term irradiance dataset. Satellite-derived GHI carries a typical uncertainty of ±3.5% at the annual level. DNI uncertainty is higher — often ±6 to 8% — because beam radiation is more sensitive to aerosol content and cloud microphysics that satellites struggle to resolve.

On-site measurements compress this number significantly. A 12-month pyranometer campaign at the project site, used to bias-correct a satellite dataset, can reduce GHI uncertainty to ±2 to 2.5%. A 24-month campaign brings it down further. This is why bankable yield reports for utility-scale projects typically require a measurement campaign of at least 12 months before financial close.

Simulation and Model Uncertainty

PV simulation software introduces its own uncertainty band. PVsyst, the industry reference, declares roughly ±3% model uncertainty for crystalline silicon under conditions where input parameters are well-characterized. Solargis pvPlanner, SAM, and HelioScope sit in similar ranges.

Model uncertainty grows when:

• Module performance under low irradiance is poorly characterized (a real issue for some bifacial and tandem products)
• The thermal model assumes a generic NOCT instead of validated cell-temperature behavior
• Inverter clipping curves do not match the actual inverter firmware behavior
• Tracker backtracking algorithms differ between simulation and field controller

Each gap adds uncorrelated uncertainty that compounds in the σ_total formula.

Inter-Annual Variability

Even with perfect irradiance data and a perfect model, the sun does not deliver the same energy every year. Inter-annual variability — the natural year-to-year swing in irradiance — is a real physical phenomenon driven by cloud cover, atmospheric aerosols, and large-scale weather patterns.

Typical inter-annual sigma values:

When the report quotes a P90 for a single year of operation, this full sigma flows through. When the report quotes a 10-year P90 — relevant for long-term debt sizing — the sigma is divided by √10 ≈ 3.16, which compresses the inter-annual contribution by about 70%.

Local Effects

Site-specific losses — soiling, snow, vegetation growth into the array, mismatch from manufacturing tolerance — each carry small uncertainty bands. Individually they are minor. Combined, they typically add 1 to 2% to σ_total.

Why Uncertainty Matters More Than P50

Two yield reports with identical P50 numbers can have wildly different P90s. The smaller the σ_total, the tighter the distribution, and the higher the P90. A project with disciplined irradiance measurement, validated simulation, and well-characterized losses will always have a higher bankable P90 than a project with the same P50 but loose inputs. This is why on-site measurement campaigns and high-quality simulation work pay for themselves at the financing stage.

A Worked P90 Calculation Example

Consider a 5 MWp ground-mount solar project in southern Spain. The yield assessment produces these inputs:

• Specific yield (P50): 1,680 kWh/kWp/year
• Total annual energy (P50): 5,000 kWp × 1,680 kWh/kWp = 8,400,000 kWh/year (8,400 MWh)
• GHI dataset uncertainty: 3.0%
• Model uncertainty: 3.5%
• Inter-annual variability: 2.8%
• Local losses uncertainty: 1.0%
• Module performance uncertainty: 0.8%

Step 1: Calculate combined sigma

σ_total = √(3.0² + 3.5² + 2.8² + 1.0² + 0.8²) σ_total = √(9.00 + 12.25 + 7.84 + 1.00 + 0.64) σ_total = √30.73 σ_total = 5.54% of P50

In absolute terms: 5.54% × 8,400 MWh = 465 MWh

Step 2: Apply the z-score for each P-value

Step 3: Interpret the result

The lender sizing senior debt to a 1.30x debt-service-coverage ratio at P90 will work from 7,804 MWh of saleable energy. The equity case can be modeled at P50 (8,400 MWh) for IRR analysis. The downside stress test runs P99 at 7,318 MWh — about 13% below the central forecast.

For investment decision-making, the generation and financial tool inside SurgePV runs this calculation directly from the design file, so designers and finance teams work from the same numbers throughout the project life cycle.

P50 vs P90 in Project Finance and Bankability

Bankability is the property of a project that lets it raise non-recourse debt at competitive rates. P50 and P90 are the two numbers around which the entire capital stack gets sized.

How Lenders Use P-Values

Senior debt providers look at P90, sometimes P95, and increasingly P99 for sensitivity analysis. The reasoning is simple: senior debt covenants typically require a debt-service-coverage ratio (DSCR) of 1.20x to 1.40x in every year of operation. If revenue is sized to P50, then by definition there is a 50% chance that any given year falls short of forecast. That violates the spirit of the covenant.

Standard solar project finance practice:

A properly structured solar project finance deal therefore produces three distinct revenue forecasts from the same yield report — one for the bank, one for the sponsor, and one for the tax equity investor. All three numbers come from the same underlying simulation. They differ only in which P-value they apply.

How P-Values Affect Loan Sizing

The arithmetic is direct. If a lender sizes debt to a 1.30x DSCR at P90, then maximum sustainable annual debt service equals (P90 revenue − operating cost) ÷ 1.30. A project with a tighter P50/P90 ratio supports more debt, because P90 revenue is closer to P50 revenue.

Take two projects, each with a P50 of $1.0 million in annual net revenue (after operating costs):

• Project A — P90 = 92% of P50 = $920,000 net revenue → max debt service = $707,700/year
• Project B — P90 = 88% of P50 = $880,000 net revenue → max debt service = $676,900/year

That $30,800/year delta translates into roughly $400,000 to $500,000 of additional debt capacity at typical 7% interest rates over a 20-year tenor — purely from tighter uncertainty inputs. A project that invests in 12 months of on-site irradiance measurement and a third-party-validated yield assessment routinely earns this back many times over.

Equity and PPA Implications

Equity investors model expected returns at P50. Internal rate of return calculations and dividend forecasts use the central yield estimate. Sensitivity analysis tests P75 and P90 cases to confirm equity returns hold up in below-average years.

PPA pricing negotiation typically anchors on P50 production, with pricing adjustments or cap-and-floor structures referencing P90 and P10 (the upside complement of P90). A 25-year fixed-price PPA with no production adjustment passes 100% of inter-annual variability risk to the offtaker; a P90-based take-or-pay structure passes most of that risk back to the developer. The structural choice flows directly from the yield report.

How to Read a P50/P90 Energy Yield Report

A standard bankable yield assessment runs 30 to 60 pages. The signal lives in five sections.

1. The Project Definition

This section confirms what was modeled — system size, module type, inverter configuration, tracking, racking, layout, electrical losses. The first audit step is making sure the modeled system matches the as-designed system. A mismatch of even 50 kWp on a 5 MWp project moves yield numbers by 1%.

2. The Solar Resource Section

Look for: which dataset, how many years, whether on-site measurements were used, and what the bias correction was. A satellite-only dataset with no measurement campaign produces a higher uncertainty band than a satellite-corrected dataset with 12 months of pyranometer data.

The annual GHI value should match independent reference values (PVGIS, NSRDB, NASA POWER) within 3 to 5%. Larger discrepancies need explanation.

3. The Loss Tree

A clean loss tree shows every loss component as a percentage of incident solar energy: shading, soiling, IAM, low-light, mismatch, DC wiring, inverter, AC wiring, transformer, availability. Total losses for a well-designed C&I system run 14 to 20%. For utility-scale, 13 to 17% is common. For residential, 18 to 25% is typical because of higher relative wiring losses and more complex shading geometry.

A loss tree that shows total losses below 10% is suspicious. A loss tree above 25% probably indicates over-conservatism or a poor system design.

4. The Uncertainty Table

The most important page. It should list every uncertainty contribution with a sigma value, then the combined σ_total. Cross-check that the math actually works using the root-sum-square formula. We have seen reports where the combined sigma was simply summed instead of root-sum-square, which inflates total uncertainty and produces an artificially low P90.

5. The P-Value Output Table

The final summary should show:

• P50 — annual energy in kWh and kWh/kWp
• P90 (1-year) — single-year downside
• P90 (10-year) — 10-year average downside
• P99 — extreme stress case
• Performance ratio — typically 78 to 84% for fixed-tilt, 80 to 86% for trackers

If the report only shows P50, ask for the uncertainty analysis. If it only shows P90, ask for the P50. Both numbers are essential for full project risk allocation.

Common Mistakes When Interpreting P-Values

These mistakes appear in nearly every project we audit. Each one has caused at least one project to fail covenant tests or default on a PPA in the past five years.

Mistake 1 — Treating P50 as a guarantee. P50 is a 50/50 number. Half of years produce less. Sales teams that quote P50 production in customer-facing materials without context regularly disappoint customers in below-average sun years. Use P90 for performance guarantees, P50 for expected returns.

Mistake 2 — Comparing P50s across vendors without checking uncertainty. A vendor quoting a higher P50 may simply have used looser assumptions. The right comparison is P50 alongside σ_total. The vendor with the higher P50 and looser sigma may produce a lower P90 than the competitor.

Mistake 3 — Ignoring the difference between 1-year and 10-year P90. A 1-year P90 is the production level exceeded in 9 out of 10 individual years. A 10-year P90 is the production level exceeded in 9 out of 10 ten-year averages. The 10-year value is always closer to P50, sometimes by 2 to 3 percentage points. Lenders sizing 20-year project finance debt typically use the 10-year P90, not the 1-year.

Mistake 4 — Using P50 for PPA pricing and then guaranteeing P50 production. If the PPA price was struck against P50 expected revenue and the developer guarantees P50 output, every year that produces below P50 (50% of years!) puts the developer in financial distress. The right contract structure either references P90 production with upside-share, or includes a production-adjustment factor.

Mistake 5 — Forgetting that P-values depend on the simulation, not reality. P50 and P90 are model outputs. They reflect the best modeling that current data and software can deliver. When the actual climate behaves outside the historical distribution — drought-driven smoke, novel cloud patterns, post-pandemic atmospheric clarity changes — actual production can fall outside the modeled P-curve entirely. The 2023 Indian monsoon delivered roughly 8% above P50 across many sites because aerosol loading dropped unexpectedly.

Mistake 6 — Confusing exceedance probability with confidence interval. P90 is not “we are 90% confident the system will produce X.” It is “the system has a 90% probability of producing at least X in a given year.” The two phrasings sound similar but mean different things. A confidence interval describes the precision of the estimate. An exceedance probability describes the variability of the outcome.

How Software Like PVsyst, Solargis, and SurgePV Generate P-Values

Different tools handle the P50/P90 calculation differently. The math is the same, but the inputs, defaults, and output detail vary.

PVsyst

The industry reference for utility-scale and commercial yield assessment. PVsyst handles P50/P90 through its “P50 — P90 evaluations” module. The user provides:

• A multi-year weather data file (15+ years recommended)
• The system definition
• Uncertainty inputs (data uncertainty, model uncertainty, system loss uncertainties)

PVsyst combines them in root-sum-square form and outputs the full P-curve. The Compare Weather Data tool inside PVsyst calculates inter-annual sigma directly from the input file. PVsyst’s documentation specifies z-scores of -1.28 for P90, -1.64 for P95, and -2.35 for P99. (See PVsyst documentation on P50–P90 evaluations.)

Solargis pvPlanner and Time-Series Tools

Solargis publishes a defined methodology for Pxx calculation that combines model uncertainty (typically ±5% at the P90 level), dataset uncertainty (typically ±3.5% for GHI), and inter-annual variability calculated from the time series. Their best-practice paper documents the full method and gives a worked example for a 1 kWp crystalline silicon system in Almería, Spain across 1994–2024 data. (See the Solargis Pxx methodology guide.)

NREL SAM (System Advisor Model)

The free NREL System Advisor Model includes a P50/P90 module documented in their 2012 technical report (NREL technical report on P50/P90 analysis). SAM uses a Monte Carlo approach by default, where annual yields from each historical year are run through the model directly to build the empirical distribution rather than assuming a Gaussian shape.

HelioScope, RatedPower, Aurora

These design platforms handle the P-value calculation by running the design through a long-term irradiance dataset and applying a configurable uncertainty band. HelioScope exposes the P50/P90 output directly inside the report builder. RatedPower (now part of Enverus) automates the calculation for utility-scale projects.

SurgePV

SurgePV’s solar design software and generation and financial tool generate P50, P90, and P99 yield bands directly from the design file. Designers do not need to export to a separate yield-assessment tool to get bankable numbers — the same model that drives the solar proposal software feeds the financial analysis. This matters operationally because it removes the version-mismatch risk between design and financial models that has caused project delays at financial close.

For more on how integrated yield calculation supports proposal generation, see our entry on performance simulation and yield assessment.

Action Items for Designers, Sales Teams, and Lenders

The same P50/P90 numbers carry different operational implications for each role on a project.

For solar designers and engineering teams:

• Always validate that the system model behind the P-values matches the as-built design at financial close. Even small mismatches (module count, inverter clipping ratio, electrical losses) shift P90 by 1 to 2%.
• Push for on-site irradiance measurement on any project above 1 MWp. The reduction in σ_total pays for the campaign within the first refinancing cycle.
• Use the same simulation tool from preliminary design through bankable yield assessment to eliminate model-mismatch uncertainty.

For sales engineers:

• Quote P90 for guaranteed performance language and P50 for expected production. Never use P50 in a contractual guarantee.
• When clients ask “what will this system produce?”, anchor on P50 with explicit reference to the P90 downside band.
• Build proposals from the same simulation model that the engineering team uses for the bankable assessment.

For project finance and banking teams:

• Cross-check the σ_total math in every yield report with root-sum-square independently. Summed sigmas inflate the apparent uncertainty and produce conservative-looking P90 numbers that are not actually conservative.
• Insist on both 1-year and 10-year P90 in every report. Use the appropriate one for the relevant covenant tenor.
• Stress test debt sizing at P95 and P99 even when senior covenants are written against P90.

For asset managers:

• Track actual annual production against the P-curve, not against P50 alone. A single year at P75 is normal. Three consecutive years at P75 is a degradation or operational issue, not weather variability.
• When actual production falls below P90, the cause is not always low irradiance. Inverter availability, soiling, vegetation growth, and module degradation rates above forecast are common culprits.

Conclusion

P50 and P90 are the language of solar project finance. They are not interchangeable, not equivalent, and not optional for serious project work.

Three concrete next steps for every team handling P-values:

• Audit your most recent yield report. Confirm that σ_total uses root-sum-square, that 1-year and 10-year P90 are both shown, and that the loss tree totals 14 to 20% for C&I or 13 to 17% for utility-scale.
• Standardize one yield calculation across design, sales, and finance. Multiple models produce multiple P50 numbers and erode trust at financial close.
• Match the right P-value to the right decision. P50 for IRR and PPA pricing, P90 for senior debt and performance guarantees, P99 for stress tests and tax-equity downside cases.

For deeper coverage of related topics, see our guide to solar irradiance components and the pv magazine explainer on P50, P90, and P99.

Frequently Asked Questions

What is the difference between P50 and P90 in solar?

P50 is the energy yield value with a 50% probability of being exceeded in any given year, so the system produces more than P50 in roughly half of all years and less in the other half. P90 is the value with a 90% probability of exceedance, meaning the system meets or beats P90 in 9 years out of 10. P90 is typically 6 to 12% lower than P50, depending on weather variability and modeling uncertainty.

Why do banks use P90 instead of P50 for solar projects?

Lenders size debt against the cash flow that almost always shows up, not the average year. P90 represents the production level that is achieved or exceeded in 9 out of 10 years, so debt service stays covered even in below-average sun years. Using P50 would mean accepting a 50% chance that revenue falls short of forecast in any given year, which violates standard debt-service-coverage covenants on solar project finance loans.

How is P90 calculated from a solar yield model?

P90 is derived from a normal distribution of expected annual yields. The formula is P90 = P50 minus 1.282 times the total uncertainty (sigma) expressed in kWh. Total uncertainty combines weather data uncertainty, simulation model uncertainty, and inter-annual variability using a root-sum-square approach. Software like PVsyst, Solargis pvPlanner, and SurgePV produce this calculation directly from multi-year irradiance datasets.

What is a typical P90 to P50 ratio for a solar project?

For a well-modeled rooftop or utility-scale PV project in a stable climate, P90 typically lands at 88 to 94% of P50. For sites with high inter-annual irradiance variability, like coastal monsoon climates or high-latitude regions, the gap can widen to 85% of P50 or lower. Smaller residential systems with limited measurement data often sit at the wider end of that range.

Is P50 or P90 more accurate?

Neither is more accurate. P50 is the most likely single-year outcome and is statistically the central forecast. P90 is the conservative downside used for risk allocation. They answer different questions: P50 answers what the system is most likely to produce, while P90 answers what the system will almost certainly exceed. Both numbers come from the same underlying simulation.

What is P99 used for in solar project finance?

P99 represents a 99% probability of exceedance and is used by lenders for stress-testing extreme downside scenarios in debt sizing. It corresponds to roughly 2.33 standard deviations below P50 and typically lands 12 to 20% below the mean estimate. Tax-equity investors and project finance banks sometimes use P99 alongside P90 to confirm that debt service holds even in a one-in-a-hundred-year low irradiance year.

How long should a solar irradiance measurement campaign run before financial close?

Industry standard is 12 months minimum, with 24 months preferred for large utility-scale projects. The on-site measurement is used to bias-correct a long-term satellite dataset, which compresses GHI uncertainty from roughly ±3.5% (satellite-only) to ±2 to 2.5% (corrected). The reduction in σ_total typically translates to a 1 to 2 percentage-point higher P90/P50 ratio, which directly affects debt-sizing capacity.

Does module degradation affect P50 and P90?

Yes. Module degradation reduces year-by-year energy output, so P50 and P90 are typically reported either as a Year-1 value or as a 25-year-lifetime average. Lenders usually want to see the full year-by-year P90 curve so they can size debt service in each year, including the degradation-adjusted late-life output. For monocrystalline silicon at 0.5%/year linear degradation, Year 25 production runs roughly 87% of Year 1.