Weather Attenuation Dataset Generation Method for Prediction-based Control of Non-Terrestrial High-Frequency Wireless Networks

High-frequency communications and space/air networks are needed to provide fast transmission rates with ubiquitous connections. The complicating factor, however, is that high frequencies are susceptible to high attenuation caused by adverse weather, such as clouds and rain. Given that weather is highly dynamic but follows predictable patterns, the use of Machine Learning to predict network state and thus proactively control connections is a promising network management method in these scenarios. To train these Machine Learning models, extensive datasets are needed. Unfortunately, existing data have either low time resolution (which does not sufficiently capture the fast changes in weather) or low space resolution (which is biased by focusing on limited locations). To address this, we propose two dataset generation methods that generate high-resolution data, which is still based on real-life measurements. Performance evaluation shows that generative artificial intelligence cannot properly capture the real-life variations in weather, but our proposed heuristics are capable of creating a dataset that follows realistic patterns and that can be used to control space/air wireless networks to improve downtime and throughput when compared to using existing datasets.