Maximising Weather Forecasting Accuracy through the Utilisation of Graph Neural Networks and Dynamic GNNs

Abstract

Weather forecasting is an essential task to tackle global climate change. Weather forecasting requires the analysis of multivariate data generated by heterogeneous meteorological sensors. These sensors comprise of ground-based sensors, radiosonde, and sensors mounted on satellites, etc., To analyze the data generated by these sensors we use Graph Neural Networks (GNNs) based weather forecasting model. GNNs are graph learning-based models which show strong empirical performance in many machine learning approaches. In this research, we investigate the performance of weather forecasting using GNNs and traditional Machine learning-based models.

Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.