Probabilistic yield terminology separates experienced solar developers from those who struggle at the lender’s desk. When a project finance team asks for your P90 figure, they are not asking for your best guess — they are asking for a statistically grounded number that quantifies downside risk. Get the framing wrong and your debt service coverage ratio collapses before the first panel ships.
This guide explains what P50, P90, and P99 mean in practice, how they are calculated, and what lenders actually do with these numbers when they size debt. It also covers the most common mistakes that cause yield reports to fail independent engineer (IE) review — and how to avoid each one.
Key Takeaway: P50 is your expected annual yield. P90 is the yield exceeded 90 % of the time — the number lenders use to size debt. The gap between them, typically 5–12 %, reflects meteorological and model uncertainty. Heaven Designs engineers both figures with documented uncertainty budgets so that independent engineers approve them on first review.
What “Exceedance Probability” Actually Means
Exceedance probability is the probability that actual annual energy production will be equal to or greater than a stated value. A P90 figure of 18.2 GWh means that in 90 out of 100 statistically modelled years, the plant will produce at least 18.2 GWh. In the remaining 10 years it may produce less.
The concept derives from hydrology, where engineers have long characterised river flow in terms like “100-year flood.” Solar adopted the same statistical language once it became a mainstream asset class requiring institutional project finance. The underlying mathematics is straightforward: the production distribution is modelled as approximately normal (Gaussian), and exceedance probabilities correspond to standard deviations from the mean.
The key distributions modelled are:
- Interannual meteorological variability — year-to-year fluctuation in solar resource at the site
- Model uncertainty — imprecision in PVsyst’s energy translation algorithms
- Data source uncertainty — accuracy limits of the chosen TMY dataset (Meteonorm, Solargis, NSRDB)
- Equipment degradation uncertainty — module output at end of warranty period
- Soiling uncertainty — variability in soiling rate relative to the modelled assumption
Each source of uncertainty carries a standard deviation expressed as a percentage of P50 yield. They are combined in quadrature (root-sum-of-squares) to produce the total uncertainty sigma (σ). This combined sigma is then applied to the P50 value using inverse normal distribution factors to derive P90 and P99.
Understanding this chain — from individual uncertainty sources through RSS combination to exceedance values — is essential for preparing a yield report that survives IE scrutiny.
The Three Core Exceedance Values Explained
The mathematical relationship between these values uses the inverse normal distribution. If P50 = Y and the combined one-sigma uncertainty is σ%, then:
- P90 = Y × (1 − 1.282 × σ)
- P99 = Y × (1 − 2.326 × σ)
A 7 % combined uncertainty on a P50 of 20 GWh gives P90 ≈ 18.2 GWh and P99 ≈ 16.7 GWh. The P50-to-P90 haircut is 9 % in this example. A developer who presents only P50 to a lender will find that the lender applies their own internal haircut — which is typically more conservative than the engineer’s documented P90.
How Lenders Use P90 in Debt Sizing
Project finance lenders structure debt on conservative assumptions. The standard approach uses P90 production as the base case for annual cash flow modelling, ensuring that debt can be serviced even in a below-average solar year.
| Metric | Lender Input | Typical Threshold |
|---|---|---|
| Annual energy for DSCR | P90 | 1.25× minimum DSCR |
| Tariff | PPA rate × availability factor | 95–98 % uptime |
| O&M escalation | CPI + 1 % | — |
| Module degradation | 0.45–0.60 % / year linear | — |
| Stress test energy | P99 | DSCR > 1.0× |
| Tenor | 15–20 years | — |
| Gearing | 70–80 % debt : equity | — |
A lender who receives only the P50 figure cannot size debt safely. They will either demand a haircut from their own in-house analyst (which may be more conservative than the engineer’s modelled P90) or require an independent engineer to produce the probabilistic analysis. Either outcome delays financial close.
The 1.25× minimum DSCR at P90 is not universal — some lenders in mature markets accept 1.20× for strong-credit projects, while DFIs in emerging markets sometimes require 1.30× or higher due to additional construction and operational risk. Understanding what your specific lender requires before finalising the yield report prevents late-stage surprises.
Heaven Designs structures reports with lender review in mind from day one. Every report includes a dedicated uncertainty budget table so the IE can verify assumptions without reverse-engineering the PVsyst file.
Uncertainty Budget: The Number Behind the Number
The uncertainty budget is where most yield report rejections originate. Independent engineers look for five documented uncertainty components, each with a stated value and justification.
1. Interannual Variability (IAV) This is the year-to-year spread in irradiance at a site. For a 10-year TMY dataset, IAV is typically 3–5 %. Solargis long-series datasets (20+ years) reduce IAV to 2–3 %. IAV is calculated as the standard deviation of annual GHI values from the historical record, divided by the mean. A report that states a P90 without documenting the IAV component will receive an immediate IE comment.
2. Data Source Uncertainty Each satellite dataset carries a stated GHI uncertainty. Solargis quotes ±3–4 % at 90 % confidence for India. Meteonorm v8 quotes ±4–5 %. NSRDB quotes ±3–4 % for CONUS, rising to ±5 % outside CONUS. Using multiple sources and taking the weighted average reduces this component.
3. PVsyst Model Uncertainty PVsyst’s internal model uncertainty is approximately 3–4 % (one sigma). This includes irradiance transposition error, temperature model error, and mismatch/soiling approximations. IEC 61724-3 is the standard reference for this component.
4. Soiling Uncertainty If site-specific soiling data is unavailable, applying a generic 2–3 % soiling loss carries an additional ±1–2 % uncertainty. Sites with measured soiling data from nearby plants can reduce this. High-dust sites (Rajasthan, GCC, arid Africa) should apply ±2 % minimum soiling uncertainty even with cleaning schedules.
5. Equipment Performance Uncertainty Module binning tolerance (±3 %), inverter efficiency curve accuracy (±1 %), and string sizing assumptions contribute a combined ±2–3 %. This component is reduced for projects with manufacturer power sorting guarantees or high-precision binning documentation.
Root-sum-of-squares combination of five components at typical values:
√(4² + 4² + 3.5² + 1.5² + 2.5²) = √(16 + 16 + 12.25 + 2.25 + 6.25) = √52.75 ≈ 7.3 %
This 7.3 % sigma value is what gets applied to calculate P90 and P99 from P50. On a 20 GWh P50 project, this gives P90 = 20 × (1 − 1.282 × 0.073) = 18.1 GWh.
The Heaven Designs Probabilistic Yield Framework — The 5-Layer Uncertainty Stack
Heaven Designs uses a structured five-layer approach to building defensible uncertainty budgets for every bankable yield report. This framework is applied consistently across project types and geographies.
Layer 1 — Source Triangulation We run the site with a minimum of two independent meteorological datasets. If P50 values differ by more than 3 %, we investigate the reason before selecting the reference dataset. Source uncertainty is quantified as half the inter-dataset spread, floor-limited by the dataset provider’s stated accuracy.
Layer 2 — IAV Quantification We extract long-series irradiance data (minimum 20 years where available) to calculate site-specific interannual standard deviation rather than using regional defaults. For Indian projects, Solargis provides 25-year records. This frequently reduces IAV uncertainty by 1–2 percentage points versus generic assumptions.
Layer 3 — PVsyst Loss Audit We audit every loss parameter in the PVsyst simulation for defensibility. Generic default losses (e.g., soiling at 2 %, mismatch at 2 %) are replaced with project-specific values wherever data exists. Each substitution is documented with source evidence — O&M contract soiling schedules, module flash test reports, or nearby plant performance benchmarks.
Layer 4 — Sensitivity Analysis We run ±10 % GHI sensitivity cases and ±5 % temperature coefficient sensitivity cases to confirm that P90 DSCR remains above 1.25× in stress scenarios. Results are presented in a sensitivity table. This is increasingly required by DFI lenders and shows that the project can absorb weather variability without debt service default.
Layer 5 — IE Review Readiness The report format mirrors the checklist used by leading IEs (Black & Veatch, DNV, UL). We include uncertainty budget, PVsyst simulation summary PDF, meteo comparison table, and loss breakdown — all in one document set. IE reviewers can locate every required element within minutes of receiving the package.
Comparison: One-Year vs. Ten-Year P90
Lenders sometimes request both one-year P90 and ten-year P90 values. These differ because interannual variability is averaged out over a decade.
| Horizon | IAV Component | Combined Sigma | P90 (on 20 GWh P50) |
|---|---|---|---|
| 1-year P90 | Full IAV (4.5 %) | ~7.5 % | 18.1 GWh |
| 5-year P90 | IAV / √5 = 2.0 % | ~6.0 % | 18.5 GWh |
| 10-year P90 | IAV / √10 = 1.4 % | ~5.7 % | 18.9 GWh |
| 25-year P90 | IAV / √25 = 0.9 % | ~5.4 % | 19.0 GWh |
Ten-year P90 is higher (less conservative) because meteorological variability partially cancels over a decade. Some lenders use the 10-year P90 for base-case DSCR and the 1-year P90 for minimum-year stress testing. The choice of horizon should be confirmed with the lender before finalising the yield report — requesting a change after IE submission is disruptive.
Common Errors in P50/P90 Reports
Stats: Yield Report Standards Across Markets
India-Specific Context: SECI and IREDA Requirements
Indian project finance adds layers of regulatory yield documentation. SECI tenders and IREDA loan applications increasingly require:
- Dual-source meteorological comparison (Solargis + NSRDB or Meteonorm)
- Site-specific soiling data from ground measurement or validated proxy sites within 50 km
- Degradation schedule matched to module manufacturer warranty curve
- Uncertainty budget compliant with IEC 61724-3 methodology
- P90 DSCR ≥ 1.25× confirmed against the project’s financial model
Heaven Designs has delivered bankable yield reports accepted by SECI, IREDA, and SBI Capital for projects ranging from 2 MW C&I to 100 MW utility-scale. Our reports follow IEC 61724-3 methodology and are formatted for Indian IE review timelines.
For projects targeting international DFI financing (IFC, ADB, DEG), we cross-reference our uncertainty budgets against the Equator Principles and IFC Performance Standards. DFI requirements are more detailed than domestic Indian lender requirements, particularly around IAV documentation and equipment degradation assumptions.
US Market Requirements: State and Federal Lenders
In the US market, project finance yield requirements vary by lender type:
- Commercial banks (community solar, C&I): Typically accept P90 from a recognised engineering firm without requiring an independent IE letter. Report format is less standardised.
- Tax equity investors: Require IE-reviewed yield reports with full uncertainty documentation. Goldman, JPMorgan, and other major tax equity providers have internal checklists aligned to DNV or UL methodology.
- USDA REAP grants: The programme documentation references SAM (NREL System Advisor Model) but does not prohibit PVsyst. Using PVsyst for REAP applications is accepted.
- DOE Loan Programs Office: Requires full bankable yield report with IE review letter. LPO deals are typically large (50 MW+) and follow the highest documentation standards.
Our NEC 2023 compliance guide covers the US regulatory layer for permit documentation. The yield report is a separate document set from the permit package but may be required alongside it for larger commercial or utility-scale US projects.
When to Use P99
P99 appears in three specific contexts:
- Insurance sizing: Property and business interruption insurers use P99 as the downside production scenario for premium calculation and business interruption coverage limits.
- Reserve fund sizing: Debt service reserve accounts (DSRA) are sometimes sized to cover the P99 revenue shortfall relative to annual debt service obligations.
- Equity stress testing: Equity IRR at P99 production is the extreme-downside scenario presented to board-level investment committees before final investment decision.
P99 is rarely used as the primary metric for debt sizing. If a lender requests P99 as the base case, it typically signals concern about the quality of the P90 documentation — they want to see the full uncertainty distribution rather than a single point estimate.
How Heaven Designs Delivers Lender-Ready Yield Reports
Our solar feasibility study service produces bankable P50/P90 yield reports within 10–15 business days of receiving site data. The deliverable package includes:
- PVsyst simulation files (primary and sensitivity variants)
- Meteo comparison table (minimum two datasets)
- Uncertainty budget table with component-by-component breakdown
- P50, P90, P99 annual energy estimates with statistical basis
- 10-year and 25-year production forecasts with degradation curve
- IE review readiness checklist
- Sensitivity analysis (±10 % GHI, ±5 °C temperature)
Our PVsyst simulation service covers the full simulation workflow for engineers who need the model file rather than the complete report. For developers working across multiple sites, our utility-scale solar design service includes yield report production as a standard deliverable.
We also support lender review preparation for developers who have an existing PVsyst file but need the uncertainty documentation and report formatting strengthened before IE submission. This service covers uncertainty budget construction, meteo cross-validation, and transmittal letter drafting.
For developers who need to understand how meteo source quality affects their P90, our guide on PVsyst meteo data sources explains the Meteonorm vs Solargis vs NSRDB decision in detail. For tracker projects, the PVsyst tracker yield methodology covers the additional parameters specific to SAT systems.
FAQ
What is the difference between P50 and P90 in solar energy? P50 is the median expected annual production — exceeded in 50 % of modelled years. P90 is a more conservative figure exceeded in 90 % of years. The difference represents downside protection for lenders and investors. A typical P50-to-P90 haircut is 5–12 % of annual energy.
Why do lenders require P90 rather than P50? Lenders size debt repayments against conservative production scenarios. P90 ensures that even in a below-average solar year, the project generates enough revenue to cover debt service. P50 would result in debt service shortfalls in nearly half of all years — an unacceptable risk for a long-term debt instrument.
How is P90 calculated from a PVsyst simulation? PVsyst produces a P50 (expected) annual yield. Engineers then calculate a combined uncertainty sigma (typically 6–9 %) using root-sum-of-squares combination of meteorological, model, soiling, and equipment uncertainties. P90 = P50 × (1 − 1.282 × sigma). This calculation and its inputs must be documented in a formal uncertainty budget table.
What is a typical P50-to-P90 haircut? For well-documented utility-scale sites with long-series meteo data and dual-source cross-validation, the haircut is 5–8 %. Poorly documented sites with single-source meteo data and no soiling measurements can see haircuts of 10–15 %, which may impair project bankability by pushing DSCR below minimum thresholds.
What is P99 used for in solar finance? P99 represents the production floor exceeded in 99 % of years. It is used for insurance sizing, debt service reserve calculations, and extreme downside equity stress tests. It is rarely the primary metric for debt sizing but is increasingly required by DFIs as a supplementary disclosure.
Can P90 improve with better data? Yes. Using long-series meteo data (20+ years), dual-source triangulation, and site-measured soiling data all reduce the uncertainty sigma and raise the P90 value. A 2-percentage-point reduction in sigma on a 20 GWh project recovers approximately 400 MWh of bankable production — worth hundreds of thousands of dollars in additional debt capacity.
How long does it take to produce a bankable P50/P90 report? Heaven Designs delivers yield reports with full uncertainty documentation in 10–15 business days for utility-scale projects with complete site data. Complex sites with shading, floating structures, or agrivoltaic configurations may require 15–20 business days for the additional simulation work.
What happens if my P90 DSCR is below 1.25×? Options include: increase DC capacity, negotiate a higher PPA tariff, reduce debt gearing, extend debt tenor, or reduce O&M cost assumptions with contractual backup. The yield report engineer cannot adjust numbers to fix a marginal DSCR — the underlying project economics must change to support the target debt structure.
