Direct Normal Irradiance (DNI) Prediction

DNI is disproportionately attenuated by atmospheric aerosols compared to GHI. Prediction systems must ingest real-time Aerosol Optical Depth (AOD) data from sources like CAMS or MODIS to accurately model the scattering of direct beam radiation. A failure to account for dust, smoke, or sulfate aerosols can lead to DNI forecast errors exceeding 50% during high-turbidity events, making aerosol modeling the single most critical differentiator from standard irradiance forecasting.

Pyrheliometers used for DNI measurement accept a finite solid angle of the sky (typically a 2.5° half-angle), meaning they capture not only the solar disk but also circumsolar radiation scattered by thin cirrus clouds. Advanced DNI prediction models must explicitly resolve this forward-scattering peak using delta-Eddington or two-stream radiative transfer approximations to avoid systematic overestimation bias when thin cloud layers are present.

Unlike GHI, which can still receive significant diffuse radiation under broken clouds, DNI drops to near zero the instant a cloud obscures the solar disk. This binary behavior demands high-resolution Cloud Optical Depth (COD) retrieval from geostationary satellites like GOES-R or Meteosat. Prediction systems use Cloud Motion Vectors (CMV) to advect COD fields, forecasting precise occultation timing for CSP plant ramp-rate management.

DNI is defined on a plane always perpendicular to the sun's rays, requiring the prediction system to integrate solar geometry calculations (azimuth and zenith angles) into the forecast pipeline. For CSP parabolic troughs and power towers, the model must output irradiance aligned with the collector's instantaneous tracking vector, not a fixed surface. This geometric coupling means DNI forecasts are inherently tied to solar position algorithms like the NREL SPA.

While GHI is often derived from NWP cloud fraction via empirical decomposition models, DNI benefits from NWP configurations that output the direct beam component as a prognostic variable. Models like WRF-Solar and ECMWF's IFS explicitly parameterize the direct radiative flux, accounting for three-dimensional cloud effects and sub-grid-scale variability that simple post-processing cannot recover. Direct assimilation of DNI observations further constrains the model's atmospheric state estimation.

A power tower's heliostat field focuses reflected DNI from a large spatial footprint onto a single receiver. Prediction systems must therefore forecast the spatially averaged DNI across the entire heliostat field area (often several square kilometers), not just a point measurement. This requires geostatistical upscaling techniques or convolutional neural networks that learn the spatial coherence structure of the DNI field to predict the aggregate power input to the receiver.

A normalization metric defined as the ratio of actual DNI to the theoretical DNI under cloudless, aerosol-free conditions. It isolates cloud-driven attenuation from the deterministic diurnal cycle.

• Value of 1.0: Perfectly clear sky.
• Value near 0.0: Thick cloud completely occluding the solar disk.
• Value > 1.0: Cloud enhancement events where reflections from cumulus cloud edges temporarily boost irradiance beyond clear-sky levels. The clear sky DNI baseline is typically computed using a radiative transfer model such as REST2 or McClear.

A dimensionless measure of the extinction of direct beam radiation by atmospheric aerosols such as dust, sulfate, and sea salt. AOD is the single most influential clear-sky variable for DNI accuracy.

• Measured at 550 nm by the Aerosol Robotic Network (AERONET).
• High AOD events (e.g., Saharan dust outbreaks) can reduce DNI by 30-50% even under cloudless skies.
• Operational DNI models ingest AOD forecasts from the Copernicus Atmosphere Monitoring Service (CAMS) to account for this attenuation.

A technique that derives wind velocity fields by tracking the displacement of cloud features in consecutive geostationary satellite imagery frames. For DNI forecasting, CMV is used to advect cloud fields forward in time for intra-hour horizons (0-60 minutes).

• Relies on cross-correlation algorithms applied to visible and infrared channels.
• Critical for predicting sudden DNI ramps caused by cumulus cloud passages over concentrating solar power towers.
• Often fused with NWP output in a blended nowcasting system to cover the 15-minute to 6-hour gap where NWP performs poorly.

A strictly proper scoring rule that evaluates the full predictive distribution of a probabilistic DNI forecast, not just a single deterministic value. CRPS measures the integrated squared difference between the forecast's cumulative distribution function (CDF) and the empirical observation.

• Lower CRPS is better; a perfect forecast scores 0.
• Penalizes both miscalibration (systematic bias in the spread) and lack of sharpness (overly wide prediction intervals).
• Expressed in the same units as the target variable (W/m²), making it directly interpretable for grid operators sizing operating reserves.

The solar radiation scattered by aerosols and thin cirrus clouds that appears to originate from the annular region immediately surrounding the solar disk. Standard pyrheliometers used for DNI measurement have an aperture half-angle of approximately 2.5°, meaning they accept circumsolar radiation within this field of view.

• Under high AOD or thin cirrus conditions, circumsolar radiation can constitute 10-30% of measured DNI.
• Concentrating solar power systems with high concentration ratios may not fully utilize this circumsolar component, leading to overestimation of usable thermal energy if not explicitly modeled.