Data-Driven Optimization for Minimizing the Environmental Impact of Airport Deicing Activities

This paper presents a data-driven deicing management framework to minimize the environmental impact of airport deicing activities at Dallas-Fort Worth International Airport. Airplane deicing involves the use of aircraft deicing and anti-icing fluids with a high concentration of ethylene/propylene/diethylene glycol. The glycol shearing off the airplanes during taxiing and dripping to the runway runs off and mixes with the airport receiving waters, which causes an increase in bacterial growth and a subsequent reduction in dissolved oxygen (DO) endangering the aquatic lives in the receiving waters. While reducing the amount of deicing and anti-icing fluid or carefully selecting runways might be more direct ways of reducing the impact of deicing activities on DO, the airport does not control those decisions. Rather, the only decision controlled by the airport is the assignment of airplanes to deicing pads. The choice of deicing pad is indirectly related to the choice of runway since each deicing pad had different runway probability distributions. The proposed deicing management framework uses stochastic dynamic programming (SDP) to assign airplanes in each hour to deicing pad locations, so as to maximize DO concentrations in the receiving waters, subject to airport constraints. The SDP formulation initially had over three hundred state variables consisting of water quality variables at six monitoring sites of the receiving water system, glycol usage variables for each of the airport's eight deicing pad locations, and number of airplanes deiced at each deicing pad location-runway combination. The state dimensionality was significantly reduced using a data mining process. The state transition equations in the SDP formulation were obtained using a two-phase statistical analysis based on artificially-generated and actual hourly data for the deicing activities. In particular, the hourly data were used to fit (i) decision tree models for the variables related to deicing activities, and (ii) multiple linear regression models for water quality variables and meteorological variables. The proposed deicing management framework was demonstrated using three cases during major deicing events. Improvements in DO compared with actual DO recorded in the data were mixed; however, the results show a promise in tackling such a complex real world problem.