An Enhanced New Multivariate Gwr Approach For Spatio-Temporal Pm10 Levels Prediction

Nowadays, the global awareness of climate change has led to an increasing attention in scientific literature on environmental pollution, which is defined as physical, chemical and biological alterations to the natural environment caused mainly by human activities. Among the several issues that impact on the well-being of humans and the Earth, the air quality is one of the most critical dimensions. Consequently, monitoring air pollutants is essential for supporting public policies for the safeguarding of human health and environmental sustainability. In this paper, air pollution due to particulate matter with a diameter of 10 micrometers or less (PM10), has been modelled using multivariate spatio-temporal techniques. Dierently from the previous studies, this work simultaneously captures the influence of demographic, economic, and envi- ronmental variables over time, at a highly disaggregated (municipal) spatial scale. To this end, a time-indexed geographically weighted regression (GWR) multivariate model has been applied to monthly observations of PM10 across Italian municipalities and to the covariates which most aect the level of particulate matter. This approach allows the estimation of regression coe- cients over the spatial domain for each recorded time point. Then, by computing the sample spatio-temporal variogram of the obtained coecients, the behaviour of the dependent variable with respect to each predictor has been evaluated. Furthermore, the kriging technique has been carried out to predict the regression coecients, and, consequently, to forecast the dependent variable PM10 in future time points through the multivariate GWR. In order to assess the relia- bility of the proposed procedure, the jackknife technique has been applied to a properly selected test set. This new multivariate spatio-temporal approach, based on the integration of the GWR with geostatistical techniques, improves the understanding of the complex dynamics that drive air pollution on a fine spatial scale, providing valuable insights that can inform targeted and timely environmental policies.