A1377
Title: Modeling spatial anisotropy and non-stationarity in semiparametric regression with differential penalization
Authors: Eleonora Arnone - University of Turin (Italy) [presenting]
Matteo Tomasetto - Politecnico di Milano (Italy)
Laura Sangalli - Politecnico di Milano (Italy)
Abstract: Spatial regression models are essential for analyzing environmental data and predicting phenomena that vary across space. However, conventional methods often fall short of capturing the intricate spatial dependencies and non-stationarities found in real-world datasets. The purpose is to introduce a novel parameter-cascading algorithm for spatial regression. This algorithm simultaneously estimates the unknown spatial parameters that describe anisotropy and the spatial field itself while integrating physical and domain-specific knowledge. The efficacy of the parameter-cascading algorithm is demonstrated through simulation studies and by applying it to a case study involving environmental data. Through this application, the method is shown to improve the accuracy of predictions for spatially distributed variables.
