Abstract
Crosswinds are usually a limiting factor for the normal operation of small size aircraft near an airport, and one of the first causes of weather related accidents. This paper presents a statistical downscaling model that provides a 24 h wind forecast at Reus airport (Spain).
The model is trained with 6 years historic data (1 January 2000 to 31 December 2005). The predictors are variables from two numeric weather prediction (NWP) model outputs: a mesoscale (MASS) and a global model reanalysis (GFS), at grid points near Reus airport. The predictands are the wind measured at the airport runway (__METAR__s).
The downscaling model processes the historic data, making use of several data mining tools to optimize the forecasting accuracy. Primarily, examples are separated into four clusters. Following this, a selection of the predictor variables is made with every cluster. Finally, neural networks are developed to produce a wind forecast.
The downscaling model is relatively complicated, due to the different statistical techniques used, but has better forecasting accuracy than the NWP models used and other simpler statistical downscaling models. Copyright Β© 2010 Royal Meteorological Society