Generation of calibrated probabilistic forecasts of potential airspace capacity from a high resolution ensemble for en-route flight planning

Data from the AFWA multi-model Mesoscale Ensemble Prediction System (MEPS) are translated into probabilistic en-route potential airspace capacity forecasts for experimental use in aviation planning. The MEPS is a 4 km resolution, 10 member ensemble that is run twice-a-day at 00 and 12 UTC. Vertically-integrated liquid (VIL) and echo top (ETOP) forecast data from each member of the MEPS are converted into an aviation hazard field using the weather avoidance model developed by MIT-LL. The aviation hazard field is then used to determine traffic flow blockage in the north-south and east-west directions. Blockage is estimated for pieces of airspace (e.g., sectors) by using min-cut theory to determine the degree to which air traffic flowing in each direction is blocked. Combining the capacity reduction estimates from each ensemble member, a probabilistic estimate of the potentially available capacity for a given airspace segment is obtained. The same technique is then applied to observations of VIL and ETOPs from the National Severe Storms Laboratory (NSSL) to generate the truth field. The observed capacity reduction due to the presence of weather hazards is used for calibrating the probabilistic estimates. Note that the potentially available capacity only captures the meteorological aspects of traffic management and does not include other constraints or traffic controller workload issues, which may further limit the truly available capacity. Biases in the modeled VIL and/or ETOP field can produce systematically unreliable forecasts of airspace capacity. The goal of this work is to evaluate a range of techniques for generating calibrated probabilistic airspace capacity forecasts. Using data collected during the summer of 2013 it was found that bias in both VIL and ETOPs was a function of ensemble member, valid time and forecast issue time. The reliability of the probabilistic potential airspace capacity forecasts was found to be more highly sensitive to biases in the ETOP forecast than to biases in the VIL forecast. The utility of several different bias correction techniques (e.g., Cumulative Distribution Function (CDF)-matching, logistic regression) in producing more reliable capacity forecasts is explored.

This research is supported by the NWS and through data sharing by AFWA. The views expressed are those of the authors and do not necessarily represent the official policy or position of the NWS or AFWA.