P50, P90, P99 in Solar Yield Reports — A Lender's-Eye View

A lender reviewing a solar project’s financial model has one immediate question: what is the probability that this plant generates enough energy to service its debt every year for twenty-five years? The yield report is the document that answers that question. Inside the yield report, three numbers — P50, P90, and P99 — carry almost all of the risk-allocation weight. Developers who understand how these numbers are derived can negotiate better debt terms. Developers who do not understand them get trapped in due diligence, answering IE queries for six months on a project that should have closed in two.

Direct answer. In solar yield assessment, P50 is the annual energy production with a 50% probability of exceedance — the central estimate. P90 is the yield exceeded in 90 of 100 years — the standard debt-sizing metric for DSCR calculations. P99 is the yield exceeded in 99 of 100 years — used for project survival analysis. Calculated as P90 = P50 × (1 − 1.28σ) and P99 = P50 × (1 − 2.33σ), where σ is the combined coefficient of variation from irradiance data uncertainty, modelling uncertainty, and interannual variability. The Heaven Designs Exceedance Probability Matrix maps each percentile to the debt coverage thresholds lenders apply across C&I rooftop, ground-mount utility, and DFI-financed projects.

This guide is written for Suresh — an Indian utility-scale developer preparing for IREDA or SBI financing — and for Jennifer — a US C&I developer whose term sheet requires an Independent Engineer-reviewed yield report with P50 and P90 clearly separated. By the end, you will be able to read the uncertainty table in any IE report, verify whether the P90 number is correctly derived, and identify the manipulation points that cause lenders to reject or haircut yield projections.

What Exceedance Probability Actually Means

The P-value notation is borrowed from probability statistics. P50 is the 50th percentile of the annual energy production distribution — in a long-run simulation, half of all modelled years produce more than P50, half produce less. P90 is the 10th percentile — 90% of all modelled years produce more than P90, only 10% produce less.

A critical distinction that many project developers overlook: P-values in a yield report represent one-year exceedance probabilities, not lifetime probabilities. A P90 for Year 1 means there is a 10% chance of falling below that yield in Year 1. Across a 25-year project life with independent annual production distributions, the probability of falling below P90 in at least one year is substantially higher — approximately 93%. This is why lenders use P90 for annual DSCR stress tests rather than lifetime analysis. The single-year P90 is the appropriate metric for annual debt service coverage, not for assessing whether the plant will ever underperform.

According to IRENA’s Renewable Power Generation Costs 2019 report, bankable yield assessments have become a hard prerequisite for project financing at scale — lenders in most markets now require IE-reviewed P50/P90 reports as a condition of term sheet issuance, not merely as a closing deliverable.

P50, P90, and P99 Defined with Formulas

The three exceedance probabilities derive from a single calculation chain anchored on the P50 estimate and the combined uncertainty.

Step 1: Calculate P50. The P50 yield is the output of the energy simulation — typically run in PVsyst — using best-estimate inputs for irradiance, temperature, soiling, shading, and component degradation. The result is the central estimate of annual energy yield in megawatt-hours.

Step 2: Quantify uncertainty sources. The yield report must identify and quantify each source of uncertainty as a coefficient of variation (percentage standard deviation). The primary sources are:

Uncertainty Source	Typical Range	Notes
Irradiance data (satellite)	3–5%	Depends on data provider quality and years of record
Irradiance data (ground-measured)	1–2%	Requires ≥ 10 years of calibrated data
Interannual variability	3–6%	Site-specific; higher in monsoon climates
PVsyst modelling error	2–3%	Manufacturer vs. real performance
Module degradation uncertainty	1–2%	Deviation from stated degradation rate
Soiling uncertainty	1–3%	Site-specific; higher in dusty semi-arid climates
Shading model uncertainty	1–2%	Depends on 3D model accuracy

Step 3: Combine uncertainties. Assuming independence between sources, the combined coefficient of variation is calculated using the root-sum-of-squares method:

σ_combined = √(σ_irradiance² + σ_IAV² + σ_model² + σ_degradation² + σ_soiling² + σ_shading²)

Step 4: Apply the normal distribution z-scores.

• P90 = P50 × (1 − 1.282 × σ_combined)
• P99 = P50 × (1 − 2.326 × σ_combined)

A typical Indian utility-scale project using Meteonorm or Solargis satellite data will show a combined σ of 8–10%, giving a P90/P50 ratio of 0.87–0.90 and a P99/P50 ratio of 0.77–0.81.

How Lenders Use Each Percentile — The Exceedance Probability Matrix

Different financing structures use different P-values for different purposes. The Heaven Designs Exceedance Probability Matrix maps the standard application across the project types we work on.

Project Type	P50 Use	P90 Use	P99 Use
C&I rooftop (India)	Equity IRR, equity case	DSCR base case for lender (≥ 1.2x)	Loan sweep trigger, insurance minimum
Utility-scale ground-mount (India)	Auction bid energy, promoter return	IREDA / PFC DSCR test (≥ 1.25x)	Termination event threshold
DFI-financed ground-mount (Africa)	Project revenue forecast	AfDB / IFC DSCR covenant (≥ 1.3x)	Political risk cover trigger
IPP with PPA (USA C&I)	Merchant revenue forecast	Lender P90 underwriting (≥ 1.15x)	Project failure definition for guarantees

The DSCR thresholds shown above are indicative of typical market practice. Individual term sheets will state the specific P90 DSCR covenant. Indian PSU lenders (IREDA, PFC, REC) typically require P90 DSCR of 1.25x or higher. DFI lenders (AfDB, IFC) typically require 1.30x. US commercial banks for C&I PPAs range from 1.10x to 1.20x depending on offtaker credit quality.

According to NREL’s 2019 Solar PV Financing Best Practices report, P90-based DSCR underwriting has become the universal standard for project finance debt in solar, displacing earlier approaches that relied solely on P50 with a blanket haircut.

Modeling Exceedance Probabilities in PVsyst

PVsyst 7.x includes a dedicated uncertainty module that generates the P90 and P99 estimates directly from the simulation. The workflow is as follows:

1. Run the baseline simulation. Use the best-estimate inputs: the selected meteo dataset, module datasheet parameters, inverter efficiency curves, near-shading scene, and degradation rate. The output is the P50 energy yield.

2. Enter the uncertainty table. In PVsyst, navigate to the “Uncertainties” tab in the simulation parameters. The tool prompts you to enter a percentage uncertainty for each source: irradiance data, measurement / model error, soiling, interannual variability, and module performance.

3. PVsyst calculates the combined uncertainty. The software performs the root-sum-of-squares combination and applies the normal distribution z-scores to output the P75, P90, and P99 yields directly.

4. Validate the inputs. The critical step is not the PVsyst calculation — the software does the math correctly. The critical step is ensuring the uncertainty inputs reflect the actual quality of the data and model. An IE reviewer will spend most of their time interrogating the irradiance data source, the interannual variability figure, and whether the soiling loss is modeled or assumed.

Common manipulation point: developers who want to show an artificially high P90 will enter a low interannual variability figure — often 3% for sites where the real IAV is closer to 6%. This produces a P90/P50 ratio that looks optimistic on paper but does not survive IE review.

For more on bankable PVsyst report construction, including the exact uncertainty table entries that IE firms accept without pushback, see our dedicated guide.

Meteo Data Source Impact on Uncertainty Bands

The choice of meteorological data source is the single largest driver of uncertainty band width. Ground-measured data from a properly calibrated on-site pyranometer with 10+ years of record produces the tightest uncertainty bands. Satellite-derived data from commercial providers (Meteonorm, Solargis, NASA POWER) produces wider bands, with the specific width depending on the data density and validation methodology of the provider.

Meteo Data Source	Typical Irradiance Uncertainty	Typical IAV	Notes
On-site pyranometer (≥ 10 yr)	1–2%	3–5%	Best available; rarely achievable for new sites
Solargis (TMY)	3–4%	4–6%	IEC 61724 validated; accepted by most DFIs
Meteonorm 8.x	3–5%	4–6%	Strong global coverage; accepted by IREDA / PFC
NASA POWER	5–8%	5–7%	Free but wider uncertainty; use for pre-feasibility only
ERA5 reanalysis	4–6%	4–7%	Increasingly used for African sites

For Indian utility-scale projects, Solargis and Meteonorm are the two most commonly accepted sources by IREDA, PFC, and SBI. For African DFI-financed projects, AfDB and IFC have accepted both Solargis and Meteonorm, with a preference for the higher-uncertainty estimate when the two sources disagree.

According to IEA PVPS Task 13’s 2018 report on uncertainties in PV energy yield assessment, the irradiance data source accounts for roughly 50–60% of the total combined uncertainty in most bankable yield assessments. Improving the irradiance data source through validated satellite data with long records is therefore the most cost-effective step a developer can take to tighten their P90/P50 ratio.

For projects where meteo data quality is uncertain, see how we handle floating solar PVsyst setups where irradiance inputs require additional correction for water surface reflectance.

How to Read an IE Report’s Yield Section

An IE yield review typically follows a standard structure. Understanding the section headings and what they mean enables you to identify the risk flags before they become financing obstacles.

Section: Meteorological Data Review. The IE will state which data source was used, the record length, the version of the dataset, and whether it was cross-validated against an independent source or on-site measurement. A well-structured yield report will show the long-term annual GHI at the site and the coefficient of variation across years — this is the IAV input.

Section: Energy Model Review. The IE will evaluate the PVsyst simulation inputs: module performance parameters (and whether they match the manufacturer datasheet used in procurement), inverter efficiency curves, DC cabling losses, transformer losses, near-shading scene accuracy, and soiling loss assumptions. Each parameter gets a status: accepted, conservative, aggressive, or requires revision.

Section: Uncertainty Analysis. This is where the P90 derivation lives. The IE will reproduce the uncertainty table and either confirm or challenge each line. Common IE challenges:

• “The interannual variability figure of 3.5% is low for this site; we are applying 5.2% based on the Solargis 20-year dataset.”
• “The module performance uncertainty of 1.5% does not account for bifacial gain model uncertainty; we are adding 1.0%.”
• “The soiling loss is assumed at 1.5% without site measurement; we are applying a 2.5% uncertainty.”

Each challenge moves the combined σ higher, which moves the P90 lower. A project that entered due diligence with a P90/P50 ratio of 0.92 may exit IE review with a ratio of 0.87 — a 5% reduction in the energy used for DSCR calculation.

Section: Conclusion / P50 and P90 Table. The IE states the accepted P50 and P90 values and whether the project passes the lender’s DSCR test at P90. If it does not pass, the IE will state what changes to the project (higher irradiance data quality, revised uncertainty inputs, or additional modules) would bring it into compliance.

The Exceedance Probability Matrix — Heaven Designs Named Framework

The Exceedance Probability Matrix is the internal tool Heaven Designs uses to pre-validate a yield report’s P90 figure before it reaches an IE reviewer. The framework maps the uncertainty inputs to the DSCR outcomes across five standard project types, allowing the engineering team to identify in advance whether the P90 yield will satisfy the lender’s covenant — and what changes to the uncertainty inputs are defensible versus what will get rejected.

The Matrix has four columns per project type: (1) the lender’s DSCR covenant at P90, (2) the maximum combined σ that still passes the covenant given the project’s P50 yield and debt structure, (3) the recommended irradiance data source to achieve that σ, and (4) the one uncertainty item most likely to be challenged by the IE for that project type.

On the next project where a lender requests a P90-based DSCR test, apply the Matrix before commissioning the yield report: identify the lender’s DSCR covenant, back-calculate the maximum allowable σ, and select the irradiance data source accordingly.

P90 in Lender Due Diligence — What the IE Reviewer Actually Checks

The IE reviewer’s job is not to re-run the PVsyst simulation. It is to verify that the inputs are defensible and that the output represents a legitimate central estimate. The review has three stages:

Stage 1: Data source verification. The IE confirms that the meteo data source used in the simulation is the one claimed in the report. Fake or substituted meteo data is rare but occurs in pressure-deadline situations. The IE will request the raw Meteonorm or Solargis export files.

Stage 2: Input parameter audit. Each PVsyst input is checked against the procurement specification. If the report uses a module Pmax of 560 W but the bill of materials shows a module rated at 545 W, the IE will flag a discrepancy. Every mismatch between the simulation input and the procurement document adds uncertainty.

Stage 3: Uncertainty table audit. The IE applies their own uncertainty estimates to each line item and recalculates the combined σ. If the IE’s σ is higher than the developer’s σ, the IE will either accept the developer’s figure with a noted reservation or formally require a revision before issuing a positive opinion.

For more detail on how lenders conduct engineering due diligence in India, see our companion guide covering the full DD process from bid stage through financial close.

Common PVsyst Errors That Distort P90 Values

The most damaging errors in PVsyst yield reports are not calculation mistakes — they are input errors that produce plausible-looking but incorrect P90 outputs. Understanding these errors protects both developers and lenders from funding projects with misstated yield.

Error 1: Using a single-year TMY without IAV correction. A Typical Meteorological Year (TMY) dataset is a synthetic average year, not a real year. Using a TMY without adding interannual variability as a separate uncertainty source underestimates the P90-P50 gap. A report that shows IAV = 0% because “TMY is already the average” is wrong.

Error 2: Bifacial gain model without validation. PVsyst includes a bifacial gain model, but the model is sensitive to ground albedo assumptions. An assumed albedo of 0.25 (common concrete or vegetation) is often used without measurement. For ground-mount projects on sandy or highly reflective terrain, actual albedo can be 0.30–0.45, and the bifacial gain can be substantially different from the modelled value. IE firms now routinely request albedo measurement data.

Error 3: Age-independent degradation. Applying a flat linear degradation rate across all years is standard practice, but it understates early-year degradation (Light-Induced Degradation and Light and Elevated Temperature Induced Degradation) for specific module types. For bifacial monocrystalline modules, the first-year LID loss of 1–2% should be separated from the steady-state degradation rate of 0.4–0.5%/year.

See our guide on common PVsyst errors that affect bankability for a complete audit checklist covering 12 input errors that cause IE rejection.

According to IEA’s Solar PV Global Supply Chains report, the variance between modelled and actual solar plant output globally averages 5–8%, with the largest contributors being irradiance data quality and soiling underestimation — two of the primary uncertainty sources in the P90 calculation.

How Heaven Designs Delivers IE-Accepted Yield Reports

Heaven Designs has delivered yield reports accepted by IREDA, PFC, SBI, AfDB, and IFC across projects ranging from 1 MW C&I rooftop to 100 MW utility-scale ground-mount. Our yield report process follows the bid-stage to IFC-stage engineering progression.

• Solar Ground Mount Design — Utility-scale PVsyst simulations with bankable uncertainty tables, meteo data validation, and bifacial gain modelling. Ready for IE submission.
• Solar Rooftop Detailed Engineering Design — C&I rooftop yield reports including P50/P90 uncertainty analysis, soiling loss documentation, and full PVsyst simulation file for IE review.
• MW-Scale PMC — Owner’s engineer support through IE review, including pre-submission audit of the uncertainty table and coordination of IE query responses.
• Download a sample deliverable — Redacted PVsyst report with uncertainty table and P50/P90 summary, IE-ready format.

Contact us to discuss your project’s yield report requirements and the meteo data sources we recommend for your specific site location.

FAQ

What is the difference between P50 and P90 in a solar yield report?

P50 is the annual energy yield with a 50% probability of exceedance — the central estimate of what the plant will produce in a given year. P90 is the annual energy yield with a 90% probability of exceedance — the conservative estimate used for debt sizing. The difference between P50 and P90 is determined by the combined uncertainty of the yield model, with typical P90/P50 ratios of 0.87–0.90 for Indian utility-scale projects using Solargis or Meteonorm data.

Why do lenders use P90 instead of P50 for debt sizing?

Lenders use P90 because it represents a conservative energy estimate with a 90% probability of being exceeded in any given year. Using P90 for DSCR calculations means the project will cover its debt service in 90 of every 100 years (in expectation). Using P50 would mean the project fails its DSCR test in roughly half of all years — an unacceptable credit risk for project finance lenders.

What is P99 used for in a yield report?

P99 is used for project survival analysis — the yield estimate that is exceeded in 99 of 100 years. Lenders and insurers use P99 to set thresholds for catastrophic underperformance events, such as loan acceleration triggers, insurance minimum production guarantees, and political risk coverage triggers. P99 is not typically used for standard DSCR calculations because it would produce a yield so conservative that most projects would not achieve viable debt terms.

How do I know if the P90 in my yield report is credible?

Check the P90/P50 ratio. For Indian utility-scale projects with Solargis or Meteonorm data, a credible P90/P50 ratio is 0.87–0.90. A ratio above 0.93 suggests the uncertainty inputs are understated. Specifically, check the interannual variability figure — it should be 4–6% for most Indian sites, not 2–3%. Also confirm that the irradiance data record covers at least 20 years and that the IAV is drawn from the actual variance in the dataset, not assumed.

What irradiance data sources do Indian DFI lenders accept?

IREDA, PFC, and SBI have accepted both Solargis and Meteonorm 8.x as irradiance data sources for project finance. AfDB and IFC have also accepted both, with occasional preference for the source showing higher uncertainty when the two disagree. NASA POWER is acceptable for pre-feasibility analysis but is generally not sufficient for bankable yield reports due to its wider uncertainty band. On-site pyranometer data with 10+ years of calibrated record is the strongest possible source but is rarely available for new project sites.

How many IE review rounds should I expect for a yield report?

A well-prepared yield report with documented uncertainty inputs typically completes IE review in one to two rounds. Poorly prepared reports with assumed inputs and undocumented IAV figures can require three to four rounds, adding four to eight weeks to the financing timeline. The single most effective step to reduce IE review rounds is to pre-document the uncertainty table justifications and submit them with the initial report package.

Can PVsyst generate P90 and P99 directly?

Yes. PVsyst 7.x includes an uncertainty module that generates P75, P90, and P99 outputs directly from the simulation, using the uncertainty inputs you provide. The software performs the root-sum-of-squares combination and applies normal distribution z-scores. However, the accuracy of the P90 output depends entirely on the quality and defensibility of the uncertainty inputs — PVsyst does the arithmetic correctly, but the inputs must reflect the actual data quality and site conditions.

What happens if the P90 DSCR fails the lender’s covenant?

If the P90 DSCR falls below the lender’s covenant threshold, there are four options: (1) improve the irradiance data quality by upgrading to a higher-quality dataset with a longer record, (2) reduce modelling uncertainty by obtaining site-specific soiling measurements or albedo measurements, (3) increase the system size (add more modules or reduce losses) to raise the P50 yield, or (4) restructure the debt to reduce annual debt service. Option 1 is the fastest and least capital-intensive in most cases.