A Method for Wind Speed Forecasting in Airports Based on Nonparametric Regression

Abstract

Wind is one of the parameters best predicted by numerical weather models, as it can be directly calculated from the physical equations of pressure that govern its movement. However, local winds are considerably affected by topography, which global numerical weather models, due to their limited resolution, are not able to reproduce. To improve the skill of numerical weather models, statistical and data analysis methods can be used. Machine learning techniques can be applied to train a model with data coming from both the model and observations in the area of interest. In this paper, a new method based on nonparametric multivariate locally weighted regression is studied for improving the forecasted wind speed of a numerical weather model. Wind direction data are used to build different regression models, as a way of accounting for the effect of surrounding topography. The use of this technique offers similar levels of accuracy for wind speed forecasts compared with other machine learning algorithms with the advantage of being more intuitive and easy to interpret.