Home » Forecast Data » SolarAnywhere® Forecast Data Model

SolarAnywhere® Forecast Data Model

The SolarAnywhere® Forecast model is built using the most advanced algorithms developed by Dr. Richard Perez at the University at Albany – State University of New York (SUNY-Albany).1 It uses a combination of two methodologies: the satellite Cloud Motion Vector (CMV) approach and stochastic blending of Numerical Weather Prediction (NWP) models.

SolarAnywhere forecasts at short time horizons, typically from 1-minute up to 5-hours ahead, are mostly generated using the CMV approach. The CMV approach uses sequential satellite images to predict future cloud location. New satellite images are obtained and processed as soon as they become available to minimize latency and improve forecast accuracy.

At forecast horizons between 5-hours and up to 7 -days ahead, the model implements an optimized blend of numerical weather prediction (NWP) models. The blend varies by location and forecast horizon. Prior to blending, NWP data—including relative humidity and cloud cover fraction—are converted into irradiance using SolarAnywhere clear-sky radiative transfer models.

Because SolarAnywhere forecasts use a blend of both satellite-based cloud motion vector forecasts and NWPs, they capture the effects of both near-term cloud cover variability as well as larger-scale, localized atmospheric trends. SolarAnywhere forecasts may be used in conjunction with SolarAnywhere energy modeling services to provide forecasted solar energy production.

The SolarAnywhere Forecast has demonstrated industry-leading accuracy. While site-specific model training and measured-data feedback is not required to achieve high accuracy, customer-provided data may be utilized to further improve forecast performance with the SolarAnywhere Advanced Forecast.

Forecasted irradiance, weather and energy data can be accessed via API or delivered via SFTP or email. Since reliable data delivery and timeliness are critical in most solar forecasting applications, guaranteed availability of >99.9% is offered.


References

1 Perez R, Schlemmer J, Kivalov S, Dise J, Keelin P, Grammatico M, Hoff T, Tuohy A. 2018. A New Version of the SUNY Solar Forecast Model: A Scalable Approach to Site-Specific Model Training . IEEE PVSC; WCPEC 7, Waikoloa, HI (USA). Link