Neural network-based microscale weather prediction using IoT sensor data in urban areas

Abstract

Accurate weather prediction at the urban microscale is crucial for risk prevention, emergency response, and informed citizen decision-making. This study aims to improve short-term weather forecasting in Bucharest using Internet of Things (IoT) sensor data and neural network models, with a focus on predicting temperature, wind speed, and wind direction, parameters that are key to managing local environmental challenges such as air pollution dispersion and fire propagation. Previous research on urban weather forecasting has primarily relied on mesoscale models and limited sensor coverage. Recent studies have demonstrated that IoT-based approaches, coupled with machine learning, can enhance spatial and temporal resolution. Building on these findings, our work integrates local sensor networks and pollution monitoring platforms to deliver improved microscale predictions. We collected and analyzed time-series data from multiple IoT-based sensor platforms located in Bucharest and surrounding areas, including the Airly network, which provides both meteorological and air quality parameters (e.g., PM10 and PM2.5). Several neural network architectures were implemented and compared to evaluating their predictive accuracy. The proposed models demonstrated promising performance, particularly in short-term wind direction prediction, enabling more accurate assessment of pollutant dispersion and fire spread dynamics. The outcomes support data-driven decision-making for urban planners, environmental agencies, and emergency management organizations. This research contributes to a scalable approach to integrating IoT data and neural network-based modeling for fine-grained urban weather prediction, offering an adaptable framework applicable to other metropolitan areas.