Comparative Analysis of Interpolation Models for IoT-Enabled RF-EMF Exposure Mapping in Urban Areas

The ubiquity of wireless communications in our daily environments has raised public concern about exposure to radio frequency (RF) electromagnetic fields (EMF). Manual collection of RF-EMF measurements is a time-consuming and labor-intensive task, while IoT-based monitoring systems still have quite limited coverage. Electromagnetic field strength maps have been employed to estimate EMF values at locations where a sensing/measuring infrastructure does not exist, inferring their values from measurements made at close distances, thus providing wider coverage in a cost-effective manner. However, such maps have only been created for indoor settings and relatively small geographical areas. The objective of this study is to identify the most optimal geospatial interpolation method to construct an electromagnetic field strength map of large-scale urban areas. We employed five different models to create an electromagnetic exposure map, with four of them being Gaussian process regression methods (Kriging), and the fifth being the classical weighted average method of nearest neighbors. The map was constructed for the city of Paris with a dataset consisting of 3632 measurements. Our results showed that Kriging, especially with the exponential model, outperformed the nearest-neighbor method in handling spatial autocorrelation in EMF data. In particular, removing outliers from the data set significantly improved all interpolation methods, reducing the mean square error (MSE) and variability.