Analog ensemble scheme for objective, short-term cloud forecasting

Analog forecasting is well established in meteorology. In recent years, investigators have begun to automate analog forecasting using modern statistical techniques. In this study, we developed a fuzzy, k-nearest neighbors (k-nn) method to identify an ensemble of historical analogs matching a set of atmospheric parameters. This ensemble was then used to make a short-term, probabilistic prediction of cloud-free conditions. Forecasts were produced in this manner for 1, 2, 3, 4 and 5 hours into the future. Analogs were discovered in a database comprised of a multi-year, half-hourly time series of atmospheric conditions. Atmospheric parameters in the database include cloud features or patterns identified in meteorological satellite imagery as well as variables extracted from data assimilation-derived Eta model analyses. Our investigation focused on six different local and regional targets spread across a variety of weather regimes within the continental United States. Performance of our scheme exceeded both basic persistence and the persistence probability for all targets, at all forecast times.