Clear Sky Index

The Clear Sky Index is formally defined as the ratio of measured Global Horizontal Irradiance (GHI) to the theoretical Clear Sky GHI at the same instant.

• Formula: CSI = GHI_measured / GHI_clear_sky
• A CSI of 1.0 indicates a perfectly cloudless atmosphere matching the theoretical model.
• Values > 1.0 often indicate cloud enhancement events where reflections from cumulus cloud edges temporarily boost irradiance beyond the clear sky baseline.
• Values < 1.0 quantify the degree of cloud-driven attenuation.

The primary engineering purpose of the CSI is to decouple cloud-driven variability from the deterministic diurnal cycle.

• Raw GHI time series exhibit a dominant seasonal and diurnal pattern that obscures the stochastic cloud signal.
• By dividing by the clear sky expectation, the CSI produces a stationary time series centered around 1.0 under clear conditions.
• This normalization allows Long Short-Term Memory (LSTM) and Temporal Convolutional Network (TCN) models to focus learning capacity exclusively on cloud dynamics rather than wasting parameters on predictable solar geometry.

CSI values exceeding 1.0 represent a critical forecasting edge case known as cloud enhancement or over-irradiance.

• Occurs when direct beam radiation passes through gaps in broken cumulus fields while simultaneously reflecting off the bright sides of adjacent clouds.
• These events can cause irradiance ramp rates that exceed the clear sky theoretical maximum by 20-40%.
• Grid-tied photovoltaic inverters must be rated to handle these transient over-power conditions, making accurate CSI forecasting essential for transient stability assessment at high solar penetration levels.

The CSI serves as the target variable for modern machine learning-based solar forecasting architectures rather than raw irradiance.

• Numerical Weather Prediction (NWP) models output predicted cloud optical depth, which is mapped to an expected CSI via a learned regression.
• Sky imaging systems track Cloud Motion Vectors (CMV) to advect CSI fields for intra-hour forecasts.
• The final operational power forecast is reconstructed by multiplying the predicted CSI by the theoretical clear sky GHI and the site-specific Photovoltaic simulation model efficiency curve.
• This modular pipeline allows independent validation of the cloud prediction component via the Continuous Ranked Probability Score (CRPS) on the CSI distribution.

The theoretical clear sky model must undergo site calibration to account for local microclimatic aerosol conditions that are not resolved in global reanalysis datasets like ERA5.

• A long-term record of clear-sky periods is identified using a STL Decomposition of the historical GHI time series to isolate the seasonal component.
• The Linke Turbidity or aerosol optical depth input to the clear sky model is tuned so that the median CSI during these identified clear periods converges to exactly 1.0.
• This calibration corrects for persistent local haze, urban aerosol plumes, or high-altitude dust that systematically bias the theoretical baseline.

The prediction of total shortwave solar radiation received on a horizontal surface, combining both direct and diffuse components. GHI is the primary input variable for estimating photovoltaic power output. Accurate GHI forecasting requires decomposing the signal into clear sky baseline and cloud-driven attenuation components.

• Measured in W/m²
• Directly proportional to PV panel output
• Requires separate treatment of clear sky and cloud effects

The rate of change of solar irradiance over time, typically measured in W/m² per minute. This metric quantifies sudden power fluctuations caused by moving clouds that threaten grid stability. The Clear Sky Index isolates cloud-driven ramps by normalizing against the theoretical clear sky envelope.

• Critical for frequency regulation
• Triggers reserve activation
• Correlated with cloud speed and type

A prediction that outputs a full probability distribution or quantile range of future irradiance rather than a single deterministic value. By combining Clear Sky Index normalization with ensemble NWP or quantile regression, forecasters can communicate cloud uncertainty to enable risk-based reserve sizing.

• Evaluated using CRPS scoring rule
• Enables dynamic operating reserve calculation
• Trained with Pinball Loss function

The reduction in photovoltaic panel conversion efficiency caused by the accumulation of dust, pollen, and debris on the glass surface. Soiling represents a systematic deviation from the ideal Clear Sky Index relationship, requiring separate degradation modeling in operational power forecasts.

• Can reduce output by 5-25% in arid regions
• Requires periodic cleaning schedules
• Modeled as a time-varying derate factor