Answer
Solar yield prediction estimates electricity production for a defined PV system, output boundary and period. It combines weather data, geometry, equipment models and operating assumptions. A useful forecast documents losses and uncertainty, distinguishes typical-year output from specific-year simulations, and states what P50 or P90 means for its time horizon. It does not guarantee actual production.
Last updated: November 2026
A solar yield prediction should identify what is being predicted before presenting a production number: the system design, weather reference, measurement boundary and operating period. A first-year net-export estimate is different from gross inverter generation or a lifetime average.
Make the yield estimate traceable to its inputs
Explore SurgePV energy-yield modeling with a project whose weather, geometry and operating assumptions you can explain.
Book a DemoChecking cost first? See the SurgePV pricing page.
Define the forecast before choosing the model
Define the system, energy boundary and period first. Then write a forecast brief that someone else can reproduce:
| Question | Record in the prediction |
|---|---|
| Which design? | Site, layout revision, direct-current (DC) capacity at standard test conditions (STC), inverter models and alternating-current (AC) limits |
| Which energy? | Inverter output, delivery-meter energy or net export after auxiliary use |
| Which period? | First operating year, named calendar year or defined multi-year period |
| Which weather reference? | Provider, dataset/version, historical years and file type |
| Which operating basis? | Availability, cleaning, curtailment, export limits and degradation assumptions |
| Which probability? | Central estimate, exceedance level, distribution and uncertainty basis |
For storage projects, separate PV generation, battery charging/discharging, self-consumption and grid export. They cannot all share the same annual-energy label.
This guide explains the modelling workflow. The energy-yield assessment services guide addresses assessor independence, due diligence and permitted reliance. Software output alone does not prove stakeholder acceptance.
Select weather data with a traceable reference period
A weather file needs more than a provider name. Check coordinates, elevation, coverage, historical period, sampling interval, timestamps, variables, quality control and validation relevant to the site. Record whether radiation is horizontal or plane-of-array; do not feed one into a model expecting the other.
The European Commission PVGIS manual explains dataset-dependent timestamps and typical meteorological year construction. A typical meteorological year (TMY) combines selected months from a historical record. It is not a prediction of the next calendar year’s weather.
Use documented site evidence to assess horizon obstruction, shading and whether ground measurements represent the array. Sensor calibration, cleaning and placement matter. A short campaign can support a comparison, but its length alone does not prove long-term accuracy.
The IEA PVPS uncertainty review distinguishes measurement practices, instruments, module properties and modelling uncertainty. It also discusses long-term resource changes. Record any proposed resource correction separately, with evidence and a reference period; do not apply a generic regional climate adjustment.
Start with a dimensionally correct screening calculation
A screening calculation combines installed capacity, sunlight on the array and an assumed performance ratio. For a simple monofacial estimate:
E = P_DC,STC × (H_POA / G_STC) × PR
- E: energy at the stated output boundary, in kWh over the chosen period.
- P_DC,STC: installed module nameplate power, in kW.
- H_POA: plane-of-array irradiation over that period, in kWh/m².
- G_STC: reference irradiance, 1 kW/m².
- PR: an assumed dimensionless performance ratio consistent with the output boundary.
Irradiance is power per area; irradiation is energy per area integrated over time. Dividing kWh/m² by kW/m² gives hours. Multiplying by installed kW gives kWh.
Equivalently, E = A × η_STC × H_POA × PR, where A is module area in m² and η_STC is module efficiency as a fraction. These forms agree because nameplate power equals area times STC efficiency times reference irradiance. Do not apply module efficiency again after using nameplate capacity.
Hypothetical screen: 100 kW × (1,700 kWh/m² ÷ 1 kW/m²) × 0.80 = 136,000 kWh. The corresponding specific yield is 1,360 kWh/kW. The irradiation and PR are invented teaching inputs, not a site forecast or a recommended PR.
Detailed modelling replaces the aggregate assumption with time-resolved equipment and loss calculations. Read the performance ratio guide when interpreting the resulting report; PR alone does not prove forecast accuracy.
Pro Tip
Label the output boundary beside every energy value. Check that the savings model uses the same boundary, especially when batteries or auxiliary loads are present.
Build the loss model without double counting
Follow the energy through the system. For each assumption, record the evidence, model setting and energy boundary it affects.
| Model stage | Inputs to check |
|---|---|
| Resource and array plane | Radiation components, orientation, horizon, transposition and tracking geometry |
| Light reaching cells | Near shading, incidence-angle response, soiling and snow where applicable |
| Module conversion | Exact electrical model, irradiance response and cell-temperature model |
| Electrical collection | Mismatch, wiring, string configuration and operating limits |
| Inverter and delivery | Load-dependent conversion, clipping, transformer losses and auxiliary consumption |
| Operating constraints | Outages, curtailment, export control and their timing |
| Future years | Explicit degradation and equipment/operating scenarios |
A DC/AC ratio above one does not by itself quantify clipping. Its energy effect depends on the power time series, inverter limits and operating conditions. Likewise, a peak efficiency rating is not an annual conversion efficiency.
PVsyst’s loss-diagram documentation describes sequential energy stages. Check the denominator of each percentage before adding losses: a loss applied after an earlier reduction is not a percentage of the original energy. Keep losses already embedded in a module, weather or efficiency model out of separate adjustment factors.
For bifacial modules and trackers, document rear-side resource, albedo, geometry, shading, control strategy and availability. A generic gain percentage cannot substitute for those inputs.
State uncertainty and the P50/P90 time horizon
P50 and P90 are exceedance values for a defined energy distribution. P50 has a modelled 50% probability of being exceeded; P90 has a modelled 90% probability. They do not promise that exactly nine of the next ten operating years will exceed a threshold.
Separate these questions in the uncertainty budget:
- How well is the long-term resource known?
- How much does weather vary between years?
- Which measurements or models have bias?
- Which equipment and operating assumptions remain uncertain?
- Are the quantities independent, correlated or scenario-based?
- Does the result describe one year or a multi-year average?
Root-sum-square combination is appropriate only where its statistical assumptions are justified. A known bias requires investigation or correction, not automatic treatment as an independent zero-mean random term. Record correlations and exclusions rather than claiming every uncertainty is covered.
PVsyst’s P50/P90 documentation describes a Gaussian annual-yield model, distinguishes long-term weather files from specific-year files, and warns against turning annual P90 into an hourly profile by uniformly scaling every interval. Review the software method and version before using its probability output.
Hypothetical probability calculation
Suppose 136,000 kWh is an assumed P50 for a defined annual-energy normal distribution, and its separately justified standard deviation is hypothetically 6% of that value. These are teaching assumptions, not observed project uncertainty.
- Standard deviation: 136,000 × 0.06 = 8,160 kWh.
- Using the approximate normal factor 1.2816: P90 = 136,000 − 1.2816 × 8,160 = 125,542.144 kWh, or approximately 125,542 kWh.
| Illustrative annual-energy quantity | kWh/year |
|---|---|
| Assumed P50 | 136,000 |
| Calculated P90 with assumed 6% standard deviation | 125,542 |
- Assumed P50136,000
- Calculated P90125,542
Source: Illustrative example, not market data; arithmetic above using the PVsyst-described normal-distribution approach, checked November 2026.
Key Takeaway
The gap between P50 and P90 depends on the justified uncertainty budget and time horizon. The 6% assumption here teaches the calculation; it is not a recommended project value.
Carry the production basis into the proposal
Walk through SurgePV generation and financial modeling to assess which inputs and assumptions your proposal needs to disclose.
Book a DemoChecking cost first? See the SurgePV pricing page.
This arithmetic does not justify the chosen distribution or uncertainty. A financing review must establish those inputs and the appropriate horizon independently. Do not present the result as guaranteed production, a default uncertainty benchmark or proof of bankability.
Validate predictions against the right comparison
After commissioning, retain the original forecast and investigate differences before recalibrating it. Compare measured production with a model using the observed weather and the same equipment, period and output boundary. A cloudy year compared with a long-term typical-year estimate cannot alone identify an equipment fault.
The Sandia PVPMC validation procedure starts with aligned measured inputs and performance data, compares modelled output, and analyses residuals. In practice, review data gaps, sensor quality, curtailment and outages alongside the energy comparison. Preserve revisions so calibration does not erase evidence of an earlier forecasting problem.
Handoff checklist for a usable yield forecast
Deliver the weather file and provenance, layout/equipment revision, software version and settings, output boundary, monthly and annual energy, loss report, operating assumptions and uncertainty method. Record what remains provisional and which changes trigger another run.
Before moving the number into a proposal, verify the same period and energy boundary are used in the savings calculation. Ask the reviewer to approve the production basis separately from tariff, price and financial assumptions.
Evaluate SurgePV using a representative project and request a walkthrough of the relevant inputs, output boundaries and exportable evidence. Check the current software’s supported weather, loss and uncertainty methods before relying on it for the required deliverable.
Frequently Asked Questions
What is solar yield prediction?
Solar yield prediction estimates electricity production for a defined PV design, output boundary and period. It combines weather, equipment, geometry and operating assumptions, with documented losses and uncertainty.
How accurate are solar yield predictions?
There is no universal accuracy percentage. Evaluate the resource evidence, model validation, operating assumptions and uncertainty for the specific output boundary and period. An uncertainty estimate is not the same as an observed prediction error.
What is the formula for solar yield prediction?
A screening formula is E = P_DC,STC × (H_POA / G_STC) × PR. Use installed kW, plane-of-array irradiation in kWh/m² and reference irradiance of 1 kW/m² to obtain kWh. The assumed PR must match the output boundary; detailed models resolve equipment and losses over time.
What is the difference between P50 and P90 yield?
P50 and P90 have modelled exceedance probabilities of 50% and 90% for a defined energy distribution and time horizon. Neither guarantees production. A one-year probability and a multi-year-average probability require different uncertainty treatment.
Which data sources are best for solar yield prediction?
Choose a documented dataset appropriate to the site, historical period, model inputs and required decision. Compare coverage, validation, timestamps and measurement evidence rather than ranking providers by name or assuming finer resolution guarantees accuracy.
What are the most common solar yield prediction mistakes?
Common mistakes include inconsistent units or meter boundaries, untraceable weather files, double-counted losses, unjustified bias corrections and probability labels without a time horizon. Comparing a typical-year forecast with measured output without accounting for observed weather can also misdiagnose performance.
How can I improve solar yield prediction accuracy?
Investigate input quality and model residuals, validate assumptions against representative measurements, and document corrections and remaining uncertainty. Preserve the original forecast and reproduce revisions. A lower uncertainty claim needs evidence; software automation alone does not prove it.
Sources
Primary research and reference material used for this desk-research article.
Where this fits
This article is part of SurgePV's Solar Design hub, which works through the topic from first principles to the decisions a project team actually has to make.
