LVP conditions at Mohamed V airport, Morocco: Local characteristics and prediction using neural networks

Abstract

Low visibility and/or ceiling conditions have a strong impact on airports' traffic and their prediction is still a challenge for meteorologists. In this paper, the local characteristics of Low Visibility Procedure (LVP) conditions are investigated and the artificial neural network (ANN) based on resilient backpropagation as supervised learning algorithm is used to predict such meteorological conditions at Mohamed V international airport, Casablanca, Morocco. This article aims to assess the ANN ability to provide accurate prediction of such events using the meteorological parameters from the Automated Weather Observation Station (AWOS) over the period from January 2009 to March 2015. First, LVP conditions were classified according to their classes (fog LVP and no fog LVP) and their sources (Runway Visual Range -RVR LVP-, Ceiling -HCB LVP- or both) for both runway end points (35R and 17L). It is found that most of LVP conditions are associated with fog and are often due to decreasing of RVR below 600m. Next, Eleven ANNs were developed to produce LVP prediction for consecutive hourly valid forecast times covering the night and early morning. The Multi-Layer Perceptron (MLP) architecture with one hidden layer is used in this study. Results show that ANNs are able to well predict the LVP conditions and are robust to errors in input parameters for a relative error below 10%. Furthermore, it is found that the ANN's skill is less sensitive to LVP type being predicted.