Forecasting solar irradiance variability using the analog method

Current solar irradiance forecasting techniques commonly rely on numerical weather prediction (NWP) for intra-day to day-ahead forecasts. Generally, NWP are spatially too coarse to accurately predict the variability statistics required for effective grid integration. For instance, the smallest cloud sizes resolvable by the North American Mesoscale Model (NAM) are on the order of 20 km. Therefore, for cloud-level wind speeds of 5 m s
^-1
, the shortest variability time-scale resolvable is approximately 1 hour. However, a significant portion of all variability occurs in time-spans of less than 1 hour. In order to resolve very short-term variability (<;5 Min.), NWP grid discretization size would need to be decreased to less than 500 m, greatly increasing computational cost. To address this, this study uses an analog downscaling technique to create a variability forecast using from a coarse 10-km NWP forecast. Overall, it is shown that the analog method effectively captures sub-grid scale variability better than using a high-resolution NWP alone.