Predictability of Fog Visibility with Artificial Neural Network for Esenboga Airport

Abstract

Fog event affects air, land and sea transportation adversely by reducing visibility, thus causes economic loss. Besides, it has animportant role in construction planning. For this reason, it is very important to predict visibility before and during fog events. In thisstudy, fog visibility prediction was made with artificial neural networks and validations were made for Esenboğa Airport.Temperature, dew point temperature, pressure, wind speed and relative humidity, which are considered to be the most importantparameters for fog occurrence, were used for 2013-2015 years to train an artificial neural network. We selected only January,February, November and December months, as those are the foggiest months for Esenboğa airport. Correlation of test part wasevaluated after training. Then, whole data for 2016-2017 years (regardless of fog existence) were used for validation of the outputagain. As a result, we found a correlation value (R) of 0.80 for the test part of 2013-2015 years; R=0.41 and root mean square error(RMSE) of 2652m for all data of the 2016 year; and R = 0.53 and RMSE = 2464m for all data of the 2017 year. The error rate (R =0.80) for the test part (2013-2015) was found acceptable. However, consistencies for the years 2016 and 2017, when all data weretested regardless of fog existence were found below expectations.