| In recent years, shifts in weather patterns have become increasingly apparent, leading to a rise in the frequency and severity of severe weather-related disruptive events across the globe. These events, which can include floods, storms, heavy rain, high winds, winter storms, heavy snow, and blizzards, pose a significant threat to public health and safety, as well as having negative economic impacts on key sectors such as agriculture, critical infrastructure, and emergency management. To address this challenge, our paper proposes a multi-modal learning approach for predicting and estimating risk for disruptions, by integrating weather- related data from multiple sources, including text and sensor recordings. Through experimental evaluation on a dataset of hourly weather data from three different climates - Alaska, Nevada, and Pennsylvania - we demonstrate that our approach outperforms alternatives that rely solely on weather recordings. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.