.avif)
DNV Solar Resource Compass
Compare multiple satellite data sets side-by-side to choose the most bankable resource file for your site.
SOLARFARMER MODEL VALIDATION
Overview of validation approach
.avif)
SolarFarmer’s core modelling engine is designed to provide high-fidelity simulations of photovoltaic (PV) system performance.
It has already underpinned six comprehensive solar-project financing due-diligence projects in the United States, totalling 1,195 MW.
Developed by DNV, the model has undergone systematic validation against measured data from operational PV systems across a wide range of geographies, system types, and configurations.
This page outlines the methodology used to validate SolarFarmer’s physical model, the metrics used to evaluate accuracy, and a summary of the results observed. The focus is strictly on the modelling engine, not on the SolarFarmer desktop software interface.
How the SolarFarmer Model works
The SolarFarmer engine simulates PV-system energy yield using a fully physics-based approach. It tracks the entire energy chain from incident solar irradiance through geometry-driven shading, electrical configuration, thermal behaviour, and loss factors—to the final AC output.
Simulations can be run hourly or sub-hourly and are not restricted to standard 8,760-hour TMY files; long, high-resolution time series are supported. This granularity captures short-duration effects such as cloud transients and inverter clipping, providing the bankable detail required for complex asset configurations and lender reviews.
Modelling Component | Description |
|---|---|
Irradiance Modelling | Calculates plane-of-array irradiance using solar position algorithms and transposition models. |
Shading analysis | Uses 3D ray tracing and/or hemicube projection to simulate beam and diffuse shading, including electrical mismatch. |
Electrical modelling | Simulates PV module and inverter performance using manufacturer-specific IV curves and operating conditions. |
Thermal modelling | Estimates module temperature based on heat balance, affecting electrical output. |
Loss modelling | Includes soiling, mismatch, LID, degradation, clipping, and availability losses, all configurable. |
This comprehensive modelling framework allows SolarFarmer to replicate real-world energy production under varied operating conditions and site constraints. For more detail on the SolarFarmer model, review the SolarFarmer knowledge base.
Validation Methodology
SolarFarmer’s validation approach compares modelled energy output to measured data from operating PV systems. The process includes:
Site Selection
A diverse sample of 44 PV systems was selected, encompassing both fixed-tilt and single-axis tracking configurations, rooftop and ground-mounted systems, and a range of climatic conditions. Systems ranged from small commercial rooftops to large utility-scale plants.
Input Data
Where available, actual site configuration and layout data were used, including array geometry, inverter capacities, and electrical design. Irradiance inputs were sourced from high-quality datasets: on-site pyranometers or trusted satellite sources. Temperature and wind speed inputs were incorporated when available to support accurate thermal modelling.
Simulation and Assumptions
Each system was modelled using the SolarFarmer engine with standardised inputs and assumptions. Known site-specific factors were applied where possible. Where assumptions were necessary (e.g. cleaning frequency or albedo), conservative and industry-aligned defaults were used.
Comparison to Measured Output
Modelled output - either AC or DC depending on the measurement point - was compared to measured monthly or annual energy yield. Accuracy was evaluated using standard statistical metrics to quantify both systematic bias and random error. More detail is available at in the validation methodology overview.
Accuracy Metrics
Two primary metrics are used to evaluate model performance:
Metric | Definition | Purpose | Interpretation |
|---|---|---|---|
Bias | 100 × (Modelled – Measured) / Measured | Measures average deviation | Negative values indicate underestimation; positive values indicate overestimation. |
Root Mean Square Error (RMSE) | Square root of the average squared error | Measures typical error magnitude | Captures both systematic and random variation in predictions. |
Bias metrics identify systematic tendencies in the model, while RMSE quantifies the average magnitude of the error. Together, they offer a robust measure of model performance. More detail is available here.
Validation Results Summary
SolarFarmer was validated on 44 PV systems: 35 fixed-tilt and 9 single-axis tracking. Comparisons were made primarily using monthly energy outputs.
System Type | Number of Site | Mean Bias (%) | Standard Deviation of Bias (%) | Mean RMSE (%) |
|---|---|---|---|---|
Fixed-tilt | 35 | -0.14 | 3.8 | 5.6 |
Single-axis tracking | 9 | -1.3 | 3.3 | 4.6 |
Results indicate low bias and consistent alignment between modelled and measured outputs across a wide range of system types and climates. More detail, and the full results are available here: Full Data.
Example Validation Cases
Example 1: Fixed-Tilt System – United Kingdom
This ground-mounted system in the UK showed very close alignment between modelled and measured energy yield, with low seasonal variation in error.
Example 2: Single-Axis Tracker – United States
This utility-scale system exhibited a modest underestimation of performance. Slight discrepancies were attributed to assumptions in the tracker control logic.
Example 3: Rooftop Fixed-Tilt – India
In this case, increased error during monsoon months was observed, primarily due to uncertainty in irradiance inputs. The model still performed within acceptable bounds.
Location | United Kingdom | United States | India |
|---|---|---|---|
System Type | Fixed-tilt | Single-axis tracker | Single-axis tracker |
Bias | +0.5% | –1.1% | –2.5% |
RMSE | 4.8% | 3.9% | 6.3% |
Title | Authors/contributors | Publication | Focus Area |
|---|---|---|---|
Tackling Terrain: Custom Tracking Algorithms in PV Plants on Complex Terrain (PVPMC) | Lopez‑Lorente, Holmgren & Alderman | PVPMC Workshop, 2024 | Introduces terrain-aware tracking controls that enhance yield accuracy in irregular topographies. |
Energy Yield Modeling of Single‑Axis Tracking PV Systems on Irregular Terrain (PVSC) | Lopez‑Lorente, Neubert & Hamer | IEEE PVSC, 2024 | Quantifies performance differences for tracking systems in complex terrain; validates terrain-adjusted algorithm improvements. |
Light & Shade: Hemicube Approach for Fast Shading Calculations (EU PVSEC) | Neubert, Hamer & Lopez‑Lorente | EU PVSEC, 2023 | Demonstrates scalable hemicube-based shading with equivalent accuracy to ray tracing, improving simulation speed. |
Shading in Utility‑Scale PV: Hemicube Methodology (Solar RRL) | Neubert, Hamer & Lopez‑Lorente | Solar RRL, Oct 2023 | Validates hemicube-based shading model across utility-scale PV, significantly reducing computational load. |
A Model for Efficient Shading Evaluation Based in Hemicube Geometry (PVPMC) | Neubert, Lopez‑Lorente, Hamer & Mercer | PVPMC Workshop, 2023 | Introduces a hemicube shading evaluation tool for large-array design workflows—balanced speed and precision. |
Addendum to the 2021 Solar Energy Assessment Validation (DNV White Paper) | Mikofski & Chan | DNV, 2023 | Updates DNV’s validation with SolarFarmer; median bias reduced from ~3.1% to ~0.3–0.8%, distribution tighter and finance-ready. |
Uncertainty in Bifacial PV Systems with High Albedo Seasonality (PVSC) | Lopez‑Lorente, Neubert & Hamer | IEEE PVSC, 2023 | Evaluates variability effects due to seasonal albedo; improves bifacial prediction under dynamic surface reflectance. |
Tracker Terrain Loss – Part Two (IEEE Journal of Photovoltaics) | Leung, Mikofski, Hamer, Neubert, Parikh, Rainey & Kharait | IEEE JPV, Jan 2022 | Measures terrain-induced energy loss for tracking systems in sloped sites; informs correction factors integrated into SolarFarmer. |
The Impact of Tracking Algorithms & Time Resolution on Yield Modelling (EU PVSEC) | Neubert, Hamer, Rainey & Mikofski | EU PVSEC, 2021 | Shows that sub-hourly simulation significantly boosts tracking model fidelity, reducing modelling error in yield estimates. |
Validation of the SolarFarmer Software with Operational Data (EU PVSEC) | Neubert, Mikofski, Hamer & Rainey | EU PVSEC, 2020 | Benchmarks against measured data from operational PV plants, demonstrating tighter bias and error distribution vs PVsyst. |
Mikofski & Rainey | IEEE PVSC, 2020 | Evaluates terrain loss patterns across sites; correction models reduce tracking error for non-flat distributions. | |
Neubert, Hamer, Kharait & Mikofski | IEEE PVSC, 2020 | Analyzes bifacial energy output sensitivity at scale; confirms model stability across size ranges. | |
Mikofski, Darawali, Hamer, Neubert & Newmiller | IEEE PVSC, 2019 | Validates bifacial model under terrain mismatch and rear-side shading using field reference data. | |
Mikofski | PVPMC Workshop, 2019 | Tool demonstration of SolarFarmer’s bifacial features; showcases modeling enhancements over earlier versions. | |
SolarFarmer (beta): Accurate Modelling of Real‑World PV Systems (PVPMC) | Mikofski | PVPMC Workshop, 2018 | Early comparative validation versus legacy tools; highlights seasonal biases and areas for tool refinement. |
Accurate Performance Predictions with Shading & Submodule Mismatch (WCPEC/PVSC) | Mikofski, Lynn, Byrne, Hamer, Neubert & Newmiller | IEEE WCPEC, 2018 | Introduces sub-module level mismatch modelling for shading loss in large PV systems; improves prediction accuracy. |
