Randy Jensen, Stottler Henke Associates, Inc.; Richard Stottler, Stottler Henke Associates, Inc.; Christian Belardi, Stottler Henke Associates, Inc.
Keywords: Space Weather, Satellite Anomalies, Data Fusion, Machine Learning, Space Domain Awareness
Abstract:
Space operators faced with the challenge of sensemaking and threat assessment for a great many satellites need to understand the significance of current conditions for situational awareness. A key step is to rule out possible explanations in arriving at a characterization. For example, if a satellite begins to exhibit abnormal behavior, this could potentially be due to hostile actions by adversaries, or internal anomalies, or external forces such as space weather. The synthesis of historical space weather and satellite anomaly data can potentially help identify new correlations from past data, and help with operational use cases involving both forensics (of a current or past situation), and forecasting (of an anticipated space weather condition that may have adverse impacts on satellite operations). In a preliminary data fusion experiment, statistical machine learning methods were used to identify historical outlier space weather conditions and uncover correlations with records of known satellite anomalies.
For this experiment, 15 years of historical space weather data were collected for 6 commonly used indicators, from archived resources provided by the Space Weather Prediction Center (SWPC) at the National Oceanic and Atmospheric Administration (NOAA). The 6 indicators included F10.7cm radio flux, X-ray flux (indicator of solar flares), proton flux (energetic particles, at two energy levels), electron (energetic particles), and Kp index (planetary geomagnetic disturbance). Using the NOAA data sets for these indicators, outlier conditions were identified over the 15-year sample, by detecting deviations from the mean over a specific threshold. The initial source for spacecraft health data was a commercially available database which aggregates publicly sourced data about spacecraft over their entire lifetimes, including records of insurance claims when satellite anomalies occur. Anomaly records include supporting text with details and in some cases causal attributions. There were 151 candidate anomalies, and when synthesized with the space weather analysis, 18 had current or recent outlier space weather conditions. Among the 18 satellite anomalies with a possible space weather correlation, the existing records for 5 of them included attribution to space weather causes, but 13 did not. Although closer examination would be needed to draw conclusions about the remaining 13, this preliminary analysis illustrates the possible results from synthesizing these data sources in a model to support space domain awareness decision-making. Although these findings are the result of initial unsupervised data fusion, there is the potential for much greater insight to be derived using similar methods with higher-fidelity data, and ultimately applied in a model to support space domain awareness decision-making.
Date of Conference: September 14-17, 2021
Track: Machine Learning for SSA Applications