Geostationary (GEO) satellites orbit the Earth at the same rate that the Earth rotates, which allows them to remain over the same spot on the planet at all times. This makes them ideal for monitoring weather patterns and tracking the movement of storms. Space agencies around the world fly GEO satellites carrying visible/thermal imagers, infrared sounders, and lightning mapping instruments. These satellites are widely used by meteorologists for monitoring weather conditions and informing numerical weather prediction. Here’s a visualization of hourly data taken from NASA/NOAA and JAXA.
NASA and NOAA operate the GOES-R series satellites cover from the Atlantic to Pacific Oceans, including North and South America. The on-board advanced baseline imager (ABI) captures visible, near-infrared, and infrared imagery at resolution higher than the best regional weather models with near-global coverage. Sixteen spectral bands including infrared spectra provides consistent observations in both daytime and nighttime conditions. NOAA’s GOES-R series of satellites produce over a terabyte of raw data each day. The amount of data produced will certianly increase as new satellites are launched and existing ones are upgraded with more advanced sensors and data collection capabilities. Sensors on GOES-R satellites are capable of capturing environmental conditions of all kinds including:
- Clouds in visible and infrared wavelengths, which provide information on their formation, movement, and evolution
- Aerosols in the atmosphere, such as dust and smoke, which can have a significant impact on air quality, human health, and weather patterns
- Temperature changes in the atmosphere, which can be used to track the movement of air masses and predict the development of weather patterns
- Water vapor, which plays a significant role in the formation of clouds, precipitation, and severe weather events
- Atmospheric winds, which predict the movement of weather systems and the development of severe weather events
Machine learning has proven to be a valuable tool in processing the raw data to produce high-frequency environmental data products. The inference efficiency of GPU accelerated computing with deep learning enables low-latency predictions. Training datasets of arbitrary size can often be generated to ensure large models do not overfit and translate across domains by collocated GEO with low earth orbit (LEO) sensors and NWP outputs. Our team at Zeus AI is working to leverage advances in machine learning, including large scale generative and foundation models, to produce cutting edge weather forecasts powered by a global suite of GEO satellites.
 Duffy, K., Vandal, T. J., & Nemani, R. R. (2022). Multisensor Machine Learning to Retrieve High Spatiotemporal Resolution Land Surface Temperature. IEEE Access, 10, 89221-89231. pdf
 Duffy, K., Vandal, T. J., Wang, W., Nemani, R. R., & Ganguly, A. R. (2022). A framework for deep learning emulation of numerical models with a case study in satellite remote sensing. IEEE Transactions on Neural Networks and Learning Systems. pdf
 Vandal, T. J., Duffy, K., McCarty, W., Sewnath, A., & Nemani, R. (2022, August). Dense Feature Tracking of Atmospheric Winds with Deep Optical Flow. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (pp. 1807-1815). pdf
 Vandal, T. J., McDuff, D., Wang, W., Duffy, K., Michaelis, A., & Nemani, R. R. (2021). Spectral synthesis for geostationary satellite-to-satellite translation. IEEE Transactions on Geoscience and Remote Sensing, 60, 1-11. pdf
 Vandal, T. J., & Nemani, R. R. (2021). Temporal interpolation of geostationary satellite imagery with optical flow. IEEE Transactions on Neural Networks and Learning Systems. pdf