Statistical Forecasting of Ceiling for New York City Airspace Based on Routine Surface Observations
A brief overview of the statistical methodology of model development will be presented. It is based on non-linear regression and includes unique components for optimized predictor scaling, predictor subset selection, and the identification of synergy between predictor pairs. Initial development is focused on modeling ceilings at LaGuardia (LGA) based on a 1977-2004 archive of regional observations. A general evaluation of model performance on independent cases will be presented, including a breakdown by forecast horizon. Forecast performance will be compared to those of persistence, the operational Terminal Aerodrome Forecasts (TAF), and NWP (via the RUC and MM5). The development space will also be classified into several phenomenological types. Separate statistical models will be developed for each type and subsequently evaluated for improvements in forecast accuracy. The study will conclude with near-term plans for real-time implementation and future development.