Cloud Advection Schemes for Short-Term Satellite-Based Insolation Forecasts

Prediction of solar insolation is a problem increasingly well-suited to observationally-based techniques at shorter forecast times. Here, the cloud distribution is well characterized by satellite imagery and the evolution of this field can be approximated to first order as simply translational (i.e. neglecting the evolution of cloud morphology). In this research, geostationary satellite observations are used with the NOAA Pathfinder Atmospheres - Extended (PATMOS-x) retrieval package, a standalone radiative transfer code, and wind field data from a numerical weather prediction (NWP) model to derive short-term (0-3) hour forecasts of insolation for applications in solar power generation. Different cloud advection schemes are compared and contrasted. One simple advection scheme determines the cloud pixel location and cloud-top height, then advects the pixel forward in space/time using the cloud-top height model wind value. A more sophisticated scheme uses additional retrieval properties to classify cloud pixel groups into cohesive objects which are then advected using model wind fields appropriate to the characteristics of the cloud group. In both cases the predictive skill falls off over time unless cloud evolution properties are introduced. Both schemes provide a short-term forecast of cloud location, which, when combined with predicted solar geometry, terrain height information, and sensor geometry determine the location of cloud shadows. The advected cloud and geometry information is used initialize a radiative transfer model to forecast insolation at these shadow locations. Presented are results of satellite-derived insolation forecasts validated against the NOAA-ESRL Surface Radiation (SURFRAD) network both in terms of point verification and area-averaged statistics.