SOLAR RESOURCE ASSESSMENT VALIDATION

Solar Resource Compass (SRC) is built on a robust foundation of technical methodologies that ensure trustworthy, bankable outputs. Every irradiance selection, yield estimate, and modeling assumption within SRC is grounded in validated, published methods used across DNV’s independent engineering and due diligence practices.

This page provides transparency into the core methods underpinning SRC’s outputs, supported by both empirical validation and peer-reviewed research. It distinguishes between foundational methodologies, which are explained in detail, and supporting models, which are summarized and linked to relevant reference materials.

Representative irradiance selection: Maximum Likelihood Estimation (MLE)

Selecting a long-term representative irradiance dataset is a foundational challenge in solar energy assessment. SRC employs Maximum Likelihood Estimation (MLE) to objectively determine which dataset most accurately reflects the long-term solar resource at a given site.

Rather than choosing a single satellite dataset or relying on averages, MLE evaluates each available source based on its reported uncertainty (normalized to a 95% confidence interval). It then combines the probability distributions of the datasets to identify the irradiance value with the highest cumulative likelihood.

Example: site alpha

Using MLE, the final GHI estimate is skewed toward Source C due to its lower uncertainty, even though Sources A and B have similar values.

Example: site beta

For Site Beta, MLE selected NSRDB TMY2 as the nearest match, but the method also yielded a small boost to the GHI estimate based on composite likelihood. The method allows incorporation of multiple data sources while transparently reflecting their uncertainties.

🔗 [Whitepaper – MLE for Solar Resource Selection (IEEE PVSC 48)]

‍

Energy modeling validation: SolarFarmer vs operational data

The SolarFarmer engine, used by SRC to model energy yields, has undergone extensive validation against operational performance data from 12 utility-scale PV projects totaling over 800 MW.

Adjustments that improved performance alignment included:

• Hourly Modeling Correction (HMC)
• Terrain and slope losses
• Wind stow behavior
• Explicit modeling of curtailment and availability losses

🔗 [2021 Solar Assessment Validation – Published 2022]
🔗 [Addendum – Published 2023]

SolarFarmer is regarded as bankable, having supported six comprehensive solar-project financing due-diligence engagements totalling 1,195 MW.

‍

Hourly Modeling Corrections (HMC): Addressing sub-hourly clipping

SRC incorporates DNV’s Hourly Modeling Correction (HMC) to reduce bias caused by sub-hourly inverter clipping, especially in systems with high inverter loading ratios (ILR).

• HMC is a machine learning model trained on high-frequency SCADA data.
• It was validated against real operational performance and shown to correct underestimations due to hourly granularity.

Key studies:

• Clipping Loss Validation – ML-based corrections validated across system types.
• ILR Bias Analysis – Quantifies ILR impact on hourly clipping bias.
• Satellite Sampling Effects – Assesses satellite dataset granularity impacts.

🔗 [DNV Solar Whitepaper - HMC Clipping Loss Validation PVSC48 NREL]

🔗 [DNV Solar Whitepaper - HMC Effect of Inverter Loading Ratio on Energy Estimate Bias PVSC49]

🔗 [DNV Solar Whitepaper - HMC Effects-of-satellite-sampling-on-subhourly-modeling-errors PVSC49]

‍

Supporting models: terrain, snow, and soiling

Uneven terrain

SRC uses a trigonometric geometry-based terrain loss model integrated into SolarFarmer. It calculates cross-axis tilt impacts on trackers due to slope.

🔗 [SolarFarmer Model: Terrain Shading in Tracker Systems]

Snow losses

Monthly snow loss factors are estimated using historical snowfall data combined with loss curves derived from operational field data. SRC’s snow model is based on the Townsend Model, developed using multi-year winter observations from the SIERRA study.

🔗[Townsend et al., Photovoltaics and Snow – SIERRA Study]

Soiling & dust

SRC uses a regional soiling loss model adapted from the Kimber et al. (2006) dataset, which evaluated performance across 250+ PV systems in California and the U.S. Southwest. The model has been modified by DNV to reflect seasonal patterns and modern datasets.

🔗 [Kimber et al., The Effect of Soiling on Large Grid-Connected PV Systems]

‍

Long-term degradation & availability

Degradation

DNV's standard system-level degradation rate is -0.64%/year, based on peer-reviewed fleet studies totaling over 7 GW.

🔗 [Jordan et al., Progress in Photovoltaics, 2022]

‍

Alignment with DNV’s bankable methodologies

Every method integrated into SRC is consistent with DNV’s broader framework for:

• P50, P90, and P95 energy yield modeling
• Independent engineering for project finance
• Resource risk quantification and yield confidence intervals

This ensures that outputs from SRC are both traceable and defensible during due diligence and lender reviews.

Technical reference library