This paper discusses the performance of Radial Basis Function networks (RBF) in a problem of spatial regression of pollutants in Madrid. Specifically, the spatial regression of NO x and O 3 is considered, in such a way that, starting from a set of measuring points provided by the air quality monitoring network of Madrid, the complete surface of the pollutants in the city is obtained. This pollutant surface can be used as an initial step for modeling intra-urban pollution using land-use regression techniques for example. Also, different works has used a pollutant surface to study the patterns of pollution in different cities in the world and also to establish their air monitoring networks under mathematical criteria. The paper is focussed in analyzing the performance of RBF networks to obtain this first pollutant surface, so different RBF training algorithms are tested in this paper. Specifically, evolutionary-based RBF training algorithms are described, and compared with classical training algorithms for RBF networks with Gaussian kernels. The inclusion of meteorological variables in the RBF networks are also discussed in the paper. The experimental part of the article studies real results of the application of RBF networks to obtain a first pollutant surface of NO x and O 3, using the data of the air pollution monitoring network of Madrid and the meteorological network of the city.
Read full abstract