Why PVWatts Estimates Are Often Wrong (And What to Use Instead)
Last updated: 2026-04-08 · Solar Benchmark
Why PVWatts Estimates Are Often Wrong (And What to Use Instead)
PVWatts is a free NREL tool built for portfolio-level policy analysis. In published validation studies, it shows approximately 38% deviation from actual measured residential system performance. Physics-based models using hourly ERA5 weather data and site-specific inputs achieve 5-7% monthly deviation.
Head-to-Head Comparison Table
| Metric | PVWatts | Physics-Based Model (pvlib + ERA5) |
|---|---|---|
| Weather data source | TMY (30-year average) | ERA5 hourly reanalysis, actual calendar year |
| Data resolution | ~10 km grid | ~0.25 degree grid, hourly timesteps |
| Performance ratio assumption | Fixed at 0.78 | Calculated from panel model, inverter, and system losses |
| Soiling model | None (default assumption only) | Adjustable by region and season |
| Accuracy vs measured performance | ~38% deviation (NREL internal validation) | ~5-7% monthly deviation |
| Appropriate use case | System sizing estimates, policy portfolios | Individual system health diagnosis |
What Each Approach Measures
PVWatts uses Typical Meteorological Year (TMY) data, a synthetic dataset constructed from 30 years of historical weather. TMY represents a statistically average year, not any actual year. Any given 12-month period can differ from TMY by 5-15% in total irradiance alone. On top of that, PVWatts applies a single fixed performance ratio of 0.78 to every system, regardless of the actual panel model, inverter efficiency, wiring losses, or tilt and azimuth specifics of the installation.
Physics-based modeling with pvlib and Open-Meteo ERA5 data uses actual hourly irradiance records for your specific address. ERA5 is a global atmospheric reanalysis dataset produced by the European Centre for Medium-Range Weather Forecasts, covering each hour from 1940 to present. The model then applies your actual panel specifications from the CEC module database, your confirmed tilt and azimuth, and inverter efficiency curves to calculate expected output for each hour of the year you are benchmarking. See /resources/methodology for full model documentation.
Key Differences
Weather data vintage. PVWatts tells you what an average year looks like. If this year had a wet cloudy spring and the TMY year did not, PVWatts will show higher expected production than physics actually allows. ERA5 reanalysis shows what the weather actually did this year.
Performance ratio. PVWatts assigns every system a 0.78 performance ratio. Premium TOPCon panels with a high-efficiency inverter may achieve a real performance ratio above 0.85. An older system with soiling and degradation may sit below 0.72. Locking all systems to 0.78 guarantees the estimate is wrong for nearly every individual installation.
Soiling and degradation. PVWatts does not model real-time soiling. A system in a high-dust or high-pollen region can lose 3-6% of production annually to soiling before cleaning. PVWatts does not capture this. Physics-based models calibrated to regional soiling rates account for this loss category. Normal panel degradation runs 0.5%/year for PERC panels and under 0.3%/year for TOPCon panels manufactured after 2023, per NREL degradation studies. PVWatts uses a default assumption that may not match your panel technology.
Shading. PVWatts handles shading only through a simple monthly shading factor input. It cannot model time-of-day shading patterns from nearby trees or structures. pvlib with accurate tilt/azimuth inputs and horizon data models hourly shading geometry.
Purpose. PVWatts was designed for policymakers and utilities estimating aggregate generation across thousands of systems. At that scale, individual errors cancel out. For a single homeowner asking whether their 8 kW system is working correctly, 38% variance makes PVWatts unreliable as a diagnostic tool.
When Each Option Makes Sense
PVWatts is appropriate for:
- Pre-purchase sizing estimates to determine approximate panel count
- State and utility incentive program calculations
- Comparing system size options before installation
- Any context where ballpark accuracy (+/-30%) is acceptable
Physics-based modeling (pvlib + ERA5) is appropriate for:
- Diagnosing whether your installed system is underperforming
- Generating a monthly or annual performance benchmark
- Investigating unexplained production drops
- Any context requiring accuracy better than 10%
Frequently Asked Questions
Can I use PVWatts to check if my solar system is working?
PVWatts is useful for pre-installation sizing estimates but its approximately 38% deviation from measured performance makes it unreliable for diagnosing individual system health. A system could be producing 20% below its actual potential and still appear within PVWatts' normal error range. For health diagnosis, you need a model using actual weather data for your address and your specific panel and inverter specifications. Source: NREL internal validation studies.
Why does NREL publish PVWatts if it is 38% off?
PVWatts is accurate enough for its intended purpose: comparing policy scenarios and sizing systems before installation. At the portfolio level (thousands of systems), individual errors average out. The 38% figure describes deviation for individual systems, not portfolio averages. NREL documents PVWatts' appropriate use cases in its published methodology.
What is ERA5 and why does it matter for solar estimates?
ERA5 is a global atmospheric reanalysis dataset from the European Centre for Medium-Range Weather Forecasts. It provides hourly estimates of solar irradiance, temperature, and wind at roughly 31 km resolution, reconstructed using actual historical observations fed into a physics model. For solar benchmarking, ERA5 data tells you exactly how much sunlight reached your location during each hour of a specific calendar year, rather than a synthetic average year.
How much does panel degradation affect my PVWatts estimate?
PVWatts applies a fixed degradation assumption that does not vary by panel technology. PERC panels degrade at approximately 0.5%/year; TOPCon panels manufactured after 2023 degrade at under 0.3%/year, per NREL degradation studies. A 10-year-old system may be producing 4-5% less than a new system simply from normal degradation, and PVWatts may not reflect that accurately for your specific panel model.
Data: pvlib physics modeling + Open-Meteo ERA5 weather data | Last updated: 2026-04-08 | Solar Benchmark