How We Calculate Your Estimates: Methodology, Data Sources, and Accuracy

Why Transparency Matters

Every number you see on this site is a modeled estimate, not a measurement. We take your location, your system configuration, and a large climate dataset, then apply well-established physics to predict how much energy the system would produce over a typical year. This page explains exactly how we do it, where the data comes from, how we validated the math against real-world plant data, and -- most importantly -- where the limits of this approach lie.

Data Sources

We rely on two primary external data sources, both free and well-established in the renewable energy industry.

NASA POWER (climate input)

When you pick a location, we query the NASA POWER project for long-term monthly averages at that latitude and longitude. POWER aggregates over 40 years of satellite and reanalysis data and is widely used by researchers and PV planners. Specifically, we pull:

• ALLSKY_SFC_SW_DWN -- all-sky surface shortwave downward irradiance (kWh/m²/day), i.e. Peak Sun Hours
• T2M / T2M_MAX / T2M_MIN -- air temperature at 2m (°C)
• WS2M -- wind speed at 2m above surface (m/s)

Kaggle SCADA Datasets (validation)

To check whether our math actually predicts reality, we compare our formulas against published SCADA (supervisory control and data acquisition) recordings from real operating plants:

• Wind Turbine SCADA Dataset -- 10-minute interval readings from a 3.6 MW onshore turbine, including measured wind speed, active power output, and the manufacturer's theoretical power curve.
• Solar Power Generation Data (Plant 1) -- 15-minute interval readings of DC and AC power from an operating Indian solar plant, plus on-site irradiance and module temperature sensors.

Solar Calculations

Our solar production model follows the NREL PVWatts methodology, which is the industry-standard tool for modeled photovoltaic production. The core formula is:

“daily_energy = P_rated × PSH × tiltCorrection × azimuthCorrection × tempDerate × shadingFactor × (1 - systemLosses)”

Each term addresses a specific real-world loss mechanism:

Temperature Derating (The NOCT Model)

Solar panels lose efficiency when hot -- typically around -0.35% per °C above the 25°C rating temperature. We estimate cell temperature using the NOCT (Nominal Operating Cell Temperature) model:

“cellTemp = ambient + (NOCT - 20) × (irradiance / 0.8)”

With a typical NOCT of 45°C this simplifies to cellTemp = ambient + 25 × (irradiance / 0.8). We deliberately scale the cell heating with actual irradiance rather than applying a flat offset -- a subtle point that makes a meaningful difference in cold climates where panels stay cool and can actually outperform their STC rating.

Solar Validation Results

We tested four models against 1,634 matched daytime data points from the Kaggle Plant 1 dataset to isolate where our formula's errors come from:

Wind Calculations

Wind is trickier than solar because the power available in wind scales with the cube of wind speed -- a 10% error in wind speed becomes a 33% error in predicted power. We take this seriously with a multi-step adjustment chain.

The Wind Speed Adjustment Chain

1. Start with NASA POWER's WS2M (wind speed at 2 meters above ground, approximating open terrain).
2. Apply terrain correction -- urban/suburban areas have much lower near-surface wind than open farmland, even if the 2m reading is identical. We reduce the NASA number by up to 65% in dense urban areas.
3. Extrapolate to hub height using the power law V_hub = V_ref × (h_hub / h_ref)^α, where α is the wind shear exponent (0.14 for open, up to 0.35 for urban).
4. Apply the turbine power curve: cubic law below rated speed, capped at nameplate above rated speed, zero outside the cut-in/cut-out range.

“P = 0.5 × ρ × A × v³ × Cp × η_generator × η_system”

Where ρ is air density (adjusted for elevation), A is the rotor swept area, v is wind speed at hub height, Cp ≈ 0.40 is the rotor aerodynamic coefficient (well below the 0.593 Betz limit), and the generator and system efficiencies (0.9 × 0.85 ≈ 0.77) cover electrical conversion losses.

Wind Validation Results

We compared our power curve against every 10-minute SCADA reading from a real 3.6 MW onshore turbine across the 3-25 m/s operating range, covering tens of thousands of data points:

What We Cannot Model

Being transparent about our limits is as important as being transparent about our methods. The following factors genuinely affect real-world production but cannot be predicted from the inputs you provide:

• Soiling dynamics -- dust, pollen, and bird droppings accumulate gradually and are washed off unpredictably by rain. We assume a constant 2% soiling loss.
• Panel and turbine degradation over time -- real systems lose ~0.5%/year for solar and more variable amounts for wind. We estimate year-one production.
• Inverter clipping -- if your DC/AC ratio exceeds your inverter's capacity, peak output is clipped. Our model does not distinguish DC vs AC capacity.
• Microclimate -- a coastal cliff, a valley floor, or a rooftop heat island can deviate significantly from the nearest NASA grid point (~50 km resolution).
• Shading from neighbors, trees, chimneys -- we use a single flat shading factor; real shading is directional and time-of-day dependent.
• Net metering, feed-in tariffs, time-of-use rates -- financial calculations assume flat electricity rates unless you configure otherwise.
• Year-to-year weather variation -- a single bad year can produce 15-20% less than the climatological average.

Intended Use

If you want to reproduce or audit any of our numbers, the full formulas live in src/lib/solar-calculations.ts and src/lib/wind-calculations.ts. The validation scripts that produced the bias numbers above are in scripts/validate/. We welcome corrections, better datasets, and improved physics -- renewable energy planning works better when the math is open.