The thermospheric density values shown in this visualization are "hindcast" values predicted by the Karman ML model using input data from 100 minutes prior to the date and time selected by the user. The model is valid for altitudes from 200 to 600 km above mean sea level (AMSL).
Karman is a data driven thermospheric density estimation with machine learning. It offers support for inputs from several different sources (solar proxies, geomagnetic indices, OMNIWeb data, GOES EUV data, SOHO EUV data, POD-derived thermospheric density data and more). It contains models for both nowcasting and forecasting of thermospheric density, leveraging different ML models, including time fusion transformers.
Karman is a machine learning derived prediction which has been built with a continual learning function to better prepare it for changes in solar phenomena, however its robustness and limitations in out-of-distribution scenarios (OOD) has yet to be determined.
We invite interested colleagues to support the evolution of Karman towards an operational product by using Karman’s Benchmarking Toolbox designed to streamline the testing and verification of performance of this ML pipeline by independent third parties. The Benchmarking Toolbox provides accessible notebooks and tutorials to support verification by independent experts.
We invite users to interpret results critically.
Karman Benchmarking Toolbox:
https://spaceml-org.github.io/karman/
Karman was developed as a collaboration between the Heliophysics Division of NASA and Trillium Technologies Inc, 8668 John Hickman Pkwy, STE 301 Frisco, TX 75034
NASA Grant numbers: 80ARC018D0010 (FDL 2023), 80NSSC24M0122 (FDL 2024), NNX14AT27A (Benchmarking)
NASA Executive, Dr Lika Guhathakurta, NASA Heliophysics Division.
For more information see "Improving Thermospheric Density Predictions in Low‐Earth Orbit With Machine Learning" (2024), Acciarini, G., Brown, E., Berger, T., Guhathakurta, L., Parr, J., Bridges, C., Baydin, A., G., Space Weather, 22, 2, https://doi.org/10.1029/2023SW003652.