Forecasting Of Airport Visibility

Nowadays,there is some consensus that the airline industry is one of the most intensely competitive industry,and optimizing all the operational costs is the key factor contributing to the competence.Inaccurate visibility forecasting at the airport may unnecessarily bring about the interruption to flight operation,which will cause the airlines bear more cost.Furthermore,it may lead to fatal accidents,which can both directly affect airline cost and indirectly affect the airline reputation,which in turn,will create a selfreinforcing impact to the airline cost.For this reason,it is important to have an accurate visibility forecasting system.This thesis presents new applicable methods for forecasting the visibility at the airport.The thesis develops models using artificial neural networks to predict the likelihood that the visibility is going to belong to each category justified by the International Civil Aviation Organization(ICAO)standard,and tests the methods with the Urumqi airport in China,an airport reputable for its poor visibility and the results show significant improvement over the current forecasting system at Urumqi airport.The best models obtained from this thesis are the model with three historical time steps,which is capable of predicting the visibility of 13:00 to 19:00.It was measured to be approximately 10% overall error for the prediction of the upcoming 6-hour and following 9-hour,and approximately 20% overall error for the prediction of upcoming 7-hour and 8-hour.Moreover,the models can be calibrated with their current forecasting system.This will increase the accuracy for approximately 5% for the model prediction of 9-hour lead hour with three historical time steps or at least maintain the accuracy for the model prediction of 1-hour lead hour with three historical time steps.The work presented here will help support the human to predict the visibility more accurately and also has profound implications for future studies of forecasting in other fields.