Analog Sky Condition Forecasting Based on a k-nn Algorithm

Abstract

Abstract Very short-range, cloudy–clear sky condition forecasts are important for a variety of military, civil, and commercial activities. In this investigation, an approach based on a k-nearest neighbors (k-nn) algorithm was developed and implemented to query a historical database to identify historical analogs matching the features of a specific instance. This ensemble of analogs was then used to make a probabilistic, clear-sky condition forecast for 1, 2, 3, 4, and 5 h into the future, for local and regional target types in two geographically distinct regions within the continental United States. The analogs were identified in a database comprised of a multiyear, half-hourly time series of atmospheric features that included cloud features identified in weather satellite imagery and meteorological variables extracted or derived from data-assimilation-based model analyses generated by NCEP’s Eta Data Assimilation System. The analog forecast scheme’s performance exceeded persistence at all five forecast intervals for both target types in both regimes based on a group of metrics including the relative operating characteristic (ROC) score, sharpness, accuracy, skill, expected normalized best cost, and reliability.