Statistical forecasting of ceiling for New York City airspace based on routine surface observations
January 29, 2006
Air traffic in the United States is highly congested in its "Northeast Corridor", an area that roughly encompasses the airspace from Washington, DC to Boston. This region is frequently affected by low cloud ceiling and visibility conditions during the cool season, often in association with synoptic-scale low pressure systems. Operating under IFR (Instrument Flight Rules) for extended periods of time substantially reduces airport capacity and can cause significant delay at major airports. Anticipating transitions into and out of IFR ceiling and visibility conditions can mitigate air traffic disruption by allowing for appropriate upstream planning. For instance, an accurate forecast of the lifting of cloud ceiling out of IFR range would allow for the release of more planes upstream to take advantage of the anticipated increase in capacity. The Federal Aviation Administration (FAA), through its Aviation Weather Research Program (AWRP), is currently sponsoring the Northeast Winter Ceiling and Visibility Project (NECV). Its purpose is to provide situational awareness of current ceiling and visibility conditions in the Northeast United States in a way tailored to the needs of air traffic control (ATC), as well as to bring a number of various but complimentary technologies to bear on providing automated 0-12 hour forecasts of upcoming conditions. Methodologies currently under development include numerical weather prediction (NWP) applications, 1-dimensional column modeling, tracking of aviation-impacting cloud, and statistical forecast models (Clark 2006). This presentation describes the development of statistical forecast models for major New York City airports. The statistical forecast models use routine regional meteorological observations as predictors for future values of ceiling and visibility for selected locations. These predictors consist primarily of hourly surface observations, but upper air soundings and buoy data are available for use as well. The methodology for building the models is based on non-linear regression, with the nonlinearity entering in the spirit of Generalized Additive Models (Hastie and Tibshiriani 1990). Several innovations are introduced to aid in predictor selection and to enhance the skill and stability of the final models. Statistical models such as these have been successfully developed and used recently in an operational setting for ATC. The recently completed San Francisco (SFO) Marine Stratus Initiative (also sponsored by AWRP) features a real-time display and forecast system, which contains as one of its components a regional statistical forecast model (Wilson 2004, Clark et al. 2005). The model uses hourly surface observations from the San Francisco Bay area along with the Oakland sounding to produce regular forecasts of stratus dissipation during the warm season. The performance of this model during two years (May – October) of real-time operations is given in Table 1. The context for the marine stratus model differs from that for NECV in several important ways. In SFO, warm season stratus dissipation is a diurnal phenomenon, governed primarily by mesoscale and radiative processes in conjunction with local topography. The NECV problem is more affected by synoptic dynamics, and less by the diurnal component. This paper next provides a high-level summary of the methodology that has been developed to build these statistical forecast models followed by details of the initial NECV problem, including some discussion of the quality of the predictor data. Model accuracy can be improved by development over phenomenological partitions of the available cases; a method of partitioning the cases is described. The paper concludes with a discussion of near-term tasks.