Applying Neural Networks in Ultra-Short Term Forecasting of Visibilty

Abstract

Visibility is a relevant meteorological variable in aviation meteorology and until nowadays forecasting of this parameter has been a complicated task. Statistical models are the most applied methods for forecasters to solve this problem. Neural networks are also nonlinear statistical models, which can produce connections between the determined meteorological variables and the visibility by learning data of measures of a previous term. After the learning process neural nets are able to estimate the visibility in the short term future. In this research numerous neural networks were tested with various topology and different form of learning data base. As a result some of these forecasts of neurals networks produced better estimates, than persistent prognosis, which are known for their relatively good evaluation in ultra-short term.