Fog Monitoring through Machine Learning of Signal Attenuation Data from Microwave Links from Cellular Communication Networks

Abstract. Fog poses significant challenges in various sectors, from transportation safety to water resource management. Traditional fog detection methods rely on limited monitoring capabilities, hampering forecasting and nowcasting. This study investigates the potential of machine learning in fog classification based on microwave link signal attenuation data, utilizing existing commercial cellular communication networks. Using data from The Netherlands, the study explores machine learning models using the McFly model architecture. By incorporating multiple predictors including Received Signal Level (RSL) data, trends, and time variables, the models aim to distinguish fog from other weather phenomena. The research extends to a broader dataset from a commercial cellular network by using a reduced model and evaluates the feasibility of applying the reduced model on a larger scale. Results indicate promising prospects for machine learning in fog detection, with the Inception Time architecture showing notable accuracy in fog classification. However, challenges remain in balancing long and short-term data to align with fog evolution and reliably to distinguish fog from precipitation. Furthermore, the study suggests exploring higher-frequency telecommunication links for enhanced fog detection systems, emphasizing the need for continuous advancements in this domain.