Satellite-to-solar irradiance model
To generate these estimates, each satellite image is processed using a technique that measures cloud cover by extracting cloud indices from visible and infrared channel images. A self-calibrating feedback process adjusts for arbitrary ground surfaces that may change over time.
For both GHI and DNI measurements, the cloud indices are used to modulate physically-based radiative transfer models describing localized clear sky climatology. Clear sky climatology models use regional averages of non-cloud-based variables, such as ozone, water vapor, aerosol optical depth (AOD) and ground elevation above sea level.
Accuracy of SolarAnywhere Data can vary based on numerous environmental factors at individual sites; however, the standard errors for annual insolation at all sites are:
- Version 2.x model: 5%, 10% and 15% for GHI, DNI, and DIF respectively
- Version 3.x model: 5%, 8% and 13% for GHI, DNI, and DIF respectively.
Perez-model validation can be found in articles posted on the SUNY Atmospheric Sciences Research Center website, including:
- A New Operational Satellite-to-Irradiance Model
- Producing Satellite-Derived Irradiances in Complex Arid Terrain
- Validation of the SUNY Model in a Meteosat Environment
Clean Power Research also provides documentation to support these bankable uncertainty numbers, including:
- SUNY Satellite-to-Solar Irradiance Model Improvements
Cloud motion vector forecast model
SolarAnywhere can provide a seamless data feed extending back to January 1998 and forecasting into the future up to seven-days-ahead. To generate forecast irradiance data in the near-term, SolarAnywhere employs a cloud motion vector model.
To generate forecasts, wind vector calculations are conducted for every Standard Resolution tile using consecutive satellite images. The wind vectors are then applied to the localized cloud index to predict movement on a temporal basis as frequently as minute-by-minute. As the forecast horizon extends out past a few hours, numerical weather prediction models are intelligently blended to account for longer-horizon physical cloud phenomena.
Typical Year File Generation
SolarAnywhere typical year irradiance data is synthesized from historical data to represent a “typical” year. SolarAnywhere typical GHI and DNI year (TGY, TDY) files are generated through a two-step process:
- The average monthly irradiation over the available time-series for each type of irradiance is calculated.
- The month from the available time-series with summed irradiance closest to the average irradiance for that month is then selected for the Typical Year dataset.
Each typical year file contains 8760 hourly values representing twelve months of irradiance closest to the average values. For example, a file might contain hourly January values from the year 2000, and hourly February values from the year 2007.
This is similar to the manner in which NREL TMY files are generated with several key differences:
- SolarAnywhere Data is calculated from a satellite source instead of a ground source.
- The values from SolarAnywhere months are not smoothed during the transition between months.
- Empirical adjustments are not applied to SolarAnywhere Data.
- The SolarAnywhere average month files are available based on 100% weighting of either GHI or DNI irradiance. TMY files contain a broader weighting of various weather variables not specifically targeted to solar resource assessment.
SolarAnywhere Satellite-to-Ground Correlation
The long-term production risk of solar project development is largely based on solar resource uncertainty. Risk can be reduced by combining satellite-to-solar and surface-collected irradiance measurements that have been collected independently of one another (i.e., non-correlated measurements).
Clean Power Research has developed an advanced approach to correlate satellite-to-solar and surface-collected irradiance measurements. In this approach, data source errors are addressed and minimized by accounting for concurrent measurements from the complementary dataset. Satellite data errors can be introduced at the stage of clear sky prediction (due largely to aerosol optical depth (AOD) and water vapor discrepancies), and during cloudy periods (i.e., relative cloud cover). Surface-collected data quality is reliant on the frequency and effectiveness of calibration and cleaning procedures, and typically are GHI-based only.
The optimized correlation method addresses specific model deficiencies by separating periods of cloudy and clear, then correlating measurements within different cloud conditions. The approach has been widely used by solar project developers to fully understand and quantify project risk on solar resource.
More information about tuning ground data can be found in the paper: Reducing Solar Project Uncertainty with an Optimized Resource Assessment Tuning Methodology.