Proximity Graphs Enhance GNNs for Improved Dust Emission Forecasting

Dust storms pose significant environmental and health risks, making accurate emission forecasting crucial. Traditional methods often struggle with the complex spatiotemporal dynamics involved in these phenomena. Researchers propose an enhanced approach using Graph Neural Networks (GNNs) combined with proximity graphs like Delaunay triangulation or k-Nearest Neighbor graphs. These proximity graphs serve as input for GNNs, enabling them to better model the intricate spatial and temporal relationships in dust data. This method significantly outperforms GNNs that use random graphs for message passing, as well as Long Short-Term Memory (LSTM) models, demonstrating its effectiveness for robust and accurate dust source prediction.