A novel message passing approach to spatial air quality prediction in urban areas

Abstract

Air pollution in our cities is a very significant cause of death and worsening of the quality of life in current times. Knowing in depth this phenomenon and its details and building tools that help us to mitigate its effects can be key to a better habitability of present and future urban environments. This is precisely the purpose of this work. Using specific data from the city of Barcelona, the problem of spatial prediction of air quality at the urban level has been addressed. The objective is to obtain a resolution at street level starting from the known values of a series of detectors, dispersed throughout the city. Certain particularities of this case study have forced us to carry out novel research in this aspect. The contributions of this work are the following. Firstly, a formalization of the problem based on Markov Random Fields has been carried out, following the recommendations of recent works in this sense, in order to favor the unification of the theoretical frameworks of the Graph Signal Processing field. After this, a novel algorithm has been developed for the resolution of the problem. This algorithm is based on a message passing scheme between nodes. In the proposed method, this algorithm is combined with a Graph Neural Network that refines the obtained result for a better approximation.