Two engineers can model the same 50 MWp plant in PVsyst and report P50 yields 6% apart. The software is not the difference; the inputs are. Lenders' independent engineers routinely re-run developer models, and any input that cannot be traced to a datasheet, a measurement or a recognised standard becomes a haircut on the financing case. This checklist covers the inputs that matter most, the typical ranges, and the documents you should keep in the data room.
The aim is a simulation that is reproducible, conservative where evidence is thin, and consistent with the equipment you are actually going to procure.
1. Meteorological Data
Start with the irradiation source. Satellite-derived datasets (Solargis, Meteonorm, NASA POWER, SolarAnywhere, ERA5-based products) have different biases by region, typically 2-5% on annual GHI and larger on DNI. Best practice for projects above 20 MWp is to use at least one long-term satellite series (15-20+ years) and, where possible, correct it with 12 months of on-site pyranometer data using measure-correlate-predict (MCP) methods. Use a typical meteorological year (TMY) for P50 and the inter-annual variability (typically 2-4% standard deviation of GHI) to derive P90 and P99.
Verify that the file carries ambient temperature, wind speed and, when available, relative humidity. Hourly files are the minimum; sub-hourly files help for high DC:AC ratios where clipping is sensitive to short-term irradiance peaks.
2. Orientation, Tracker and Shading
Enter the actual tilt, azimuth and pitch. For fixed tilt, ground coverage ratio (GCR) is pitch-dependent: GCR = module slant height / pitch. For single-axis trackers, specify the rotation limits (typically ±55-60°), the backtracking algorithm and the terrain slope. The near-shading scene must use the real table layout, including row-to-row shading, and horizon profile from the actual site (not the default flat horizon). Electrical shading should be set to according to module strings for bifacial trackers with a realistic number of cell strings per module (typically three bypass diode substrings in half-cut modules, each with two half-modules in parallel).
3. Module and Inverter Files (PAN / OND)
Use manufacturer-certified PAN and OND files, matched to the exact model you intend to buy. Check these fields:
- Power tolerance: use 0/+5 W or the measured flash-test average, not the nominal maximum.
- Low-irradiance behaviour: the PAN file should reflect IEC 61853-1 matrix data where available.
- Temperature coefficient of Pmax: typical n-type TOPCon modules are −0.29 to −0.32 %/°C; PERC around −0.34 to −0.36 %/°C.
- Bifaciality factor: 70-80% for TOPCon and 80-95% for HJT, but apply it per the PAN file, not the headline.
- Inverter efficiency curve: use the European/CEC-weighted curve and the correct MPPT voltage range. Sungrow, Huawei and SMA publish OND files; verify the DC:AC ratio and the temperature derating threshold.
4. Loss Parameters That Drive the Result
The loss diagram is where analysts disagree. Use evidence-based values and document each choice:
| Parameter | Typical range | Basis / note |
|---|---|---|
| Soiling loss | 0.5-2% (arid regions up to 4-6%) | Site-specific; use monthly profile and rainfall |
| IAM loss | 1.0-2.5% | ASHRAE b0 = 0.05 or measured IAM; ARC glass differs |
| LID (first-year) | 0.5-1.5% PERC; 0.3-1% TOPCon | See manufacturer warranty and test reports |
| Module quality / mismatch | −0.5 to +0.5% / 0.8-1.2% | Use flash-list data when available |
| DC ohmic wiring | 1.0-1.5% at STC | Calculated from cable cross-sections |
| AC ohmic and transformer | 0.5-1.5% / 1-1.5% | No-load plus load losses at MV transformer |
| Unavailability | 0.5-1.5% | Per O&M contract guarantee |
For a plant with a 1.3 DC:AC ratio, check the inverter overpower loss line separately; it is an economic choice, not a defect, but it must be modelled with hourly or sub-hourly data.
5. Albedo and Bifacial Parameters
Bifacial gain is the most over-optimistic input on many models. Use measured or satellite-derived albedo: grass 0.20-0.25, concrete 0.25-0.35, white gravel or membrane 0.40-0.60, sand 0.30-0.40, fresh snow 0.60-0.80. Apply monthly albedo values, not a single annual figure. Set the rear mismatch loss (typically 5-10%) and the structure shading on the rear (2-6% for trackers with torque tubes and 4-8% for fixed tilt racking with purlins). Row height (clearance) of 1.0-1.5 m for trackers and a pitch large enough for rear irradiance are the main levers. Compare the output against the rule of thumb: realistic bifacial gain is 3-8% on fixed tilt and 5-12% on trackers over grass.
6. Electrical Design Consistency
Check that the simulated electrical design is physically valid. Confirm the string length gives Voc at minimum site temperature below the system voltage limit (1,500 Vdc) with the 1.15-1.25 safety factor required by local code or IEC 62548, and the maximum power point voltage at maximum cell temperature remains above the inverter's MPPT minimum. Cross-check the DC cable cross-section, the combiner or string-inverter layout and the maximum input current per MPPT, including the module short-circuit current with the bifacial gain factor (Isc × 1.25 × 1.25 per NEC 690.8, or per IEC 62548).
7. Uncertainty and the P-Values
A bankable report shows P50, P75, P90 and P99 for one year and for ten years. Combine the uncertainties in quadrature:
With σ_irr = 3.5%, σ_var = 2.5%, σ_model = 2.0% and the rest around 1.5% combined, σ_total is about 5.2%, so P90 = P50 × (1 − 1.282 × 0.052) = 0.933 × P50. Make sure that you do not double count: inter-annual variability applies to the one-year P90 but is reduced in the ten-year P90.
Common Errors That Reviewers Flag
Independent engineers see the same mistakes repeatedly. First, using a PAN file from an earlier module revision whose temperature coefficients or bifaciality differ from the delivered product. Second, an optimistic albedo applied uniformly across the year when snow or vegetation changes the surface. Third, ignoring terrain-induced shading on trackers, where a 5° slope in the north-south direction can add 1-2% of backtracking loss. Fourth, running the model with a 1.0 DC:AC ratio on the datasheet while the procurement team buys at 1.35, which understates clipping and the energy gain at low irradiance. Finally, neglecting degradation: first-year LID/LeTID plus 0.4-0.55% per year linear degradation (n-type) should reduce the long-term average relative to year one, and the model must carry the year-by-year energy profile used in the financial model.
A practical safeguard is a sanity check against specific yield benchmarks. At 30°N latitude with 1,900 kWh/m² GHI, a single-axis bifacial plant should land around 1,950-2,150 kWh/kWp/year (DC basis), and a fixed-tilt plant 1,550-1,700. If your result is 10% above these, find the cause before the lender's engineer does. Also compare the performance ratio: modern plants typically show a PR of 80-86% depending on climate, and values above 88% in hot climates deserve a close look at temperature and soiling assumptions.
Finally, keep the model synchronised with the as-built design. After construction, re-run the simulation with measured soiling, the actual module flash data and the real string layout so that the performance guarantee test at COD compares like with like.
8. Documentation to Keep in the Data Room
Keep the PVsyst project file, the weather file with licence, the PAN/OND files with the manufacturer certification date, the horizon file, the loss diagram PDF and a one-page input register mapping each assumption to its source. When you change the module (for example from a 580 W PERC to a 625 W TOPCon bin) re-run the whole model with the new PAN file. Econo Solar supplies certified PAN and OND files alongside quotations for LONGi, Jinko, JA Solar modules and Sungrow or Huawei inverters, which saves time during the technical due diligence. To get files matched to your design, send us your configuration.
Frequently Asked Questions
What is a typical difference between P50 and P90 in PVsyst?
For a typical single-site project with 4-6% total uncertainty, P90 is 5-8% below P50 for a one-year basis and 3-5% below for a ten-year basis. Higher irradiance uncertainty or short measurement campaigns widen the gap.
Should I use the nominal power or the flash-test power in PVsyst?
Use the flash-test average or the guaranteed positive tolerance where documented. Using nominal power with an optimistic 0/+5 W assumption without a factory flash list can overstate yield by 0.3-0.7%, and reviewers often reject it.
How do I treat bifacial gain conservatively?
Use measured monthly albedo, include rear shading and mismatch losses, and cap the result against benchmarks (3-8% fixed tilt, 5-12% trackers over grass). If no on-site data exists, run a sensitivity with the albedo 0.05 lower.