Unlike storm predictors on Earth, these algorithms predict solar storms.
A neural network machine learning algorithm is guiding four of only 28 astronauts ever to see the far side of the moon. As the Artemis II crew flies behind the lunar surface, they will enter a 40-minute communications blackout. Radio waves travel in straight lines. The mass of the moon creates a barrier to mission control, but the mission is not flying blind. On the ground, a deep learning model is actively clearing their path through the most volatile environment in the solar system.
As a data analyst and weather fanatic, the DAGGER model guiding Artemis II astronauts is especially interesting. Deep Learning Geomagnetic Perturbation, according to NASA’s Heliophysics division, is as an advanced early warning siren. The National Weather Service uses traditional physics-based models for weather forecasting, such as the Global Forecast System (GFS) and the North American Model (NAM). NASA partnered with the U.S. Geological Survey and the Department of Energy at the Frontier Development Lab to create this siren to warn of solar wind attacks.
That’s because the primary threat to the spacecraft during its blackout is solar wind. The sun continuously ejects a stream of charged particles—plasma carrying an embedded magnetic field. During solar maxima, the sun frequently unleashes Coronal Mass Ejections (CMEs) and solar flares. If a dense, highly energized wave of solar material strikes the Orion capsule, it can induce electrical currents, fry the onboard electronics, and expose the crew to lethal doses of radiation.
The math of DAGGER
Historically, space weather was predicted using Numerical Weather Prediction models. These rely on complex fluid dynamics equations that require massive computational time. By the time a supercomputer calculates the effect of a solar storm, the storm has already hit the spacecraft. DAGGER operates by analyzing continuous telemetry from early-warning satellites positioned about a million miles sunward of Earth.
The model ingests a continuous time-series matrix that looks like this:
Instead of analyzing the individual variables, it helps to look at what this matrix represents as a whole: it is a real-time snapshot of the solar wind’s speed, density, temperature, and magnetic orientation.
The objective of DAGGER is to process that incoming snapshot and predict the effect on the Earth-Moon environment. Specifically, as detailed in the Space Weather Journal, the algorithm is trained to forecast the rate of change in the horizontal magnetic field, denoted as:
When this output reaches high thresholds, it indicates a geomagnetic storm capable of inducing destructive currents is about to hit.
The deep learning architecture
To make this prediction, DAGGER employs a recurrent neural network. Specifically, it uses Long Short-Term Memory (LSTM) nodes. LSTMs are suited for forecasting because they retain a memory of previous data states. The algorithm analyzes the preceding 120 minutes of solar wind data to establish the momentum and trajectory of the incoming particles.
The researchers trained this model using decades of historical satellite records, cross-referencing past solar wind anomalies with the corresponding magnetic disruptions that occurred on Earth during historical storms. Because the neural network calculates these established mathematical relationships directly rather than simulating the physics of every particle, the processing time drops from hours to milliseconds. DAGGER ingests the satellite telemetry and produces a global forecast in less than one second.
Because the early-warning satellites are located a million miles away, the solar wind takes approximately 30 to 45 minutes to cross the distance to the Earth-Moon system. DAGGER processes the data as it crosses that million-mile mark, providing mission control with a verified 30-minute warning before the radiation spikes.
If DAGGER detects an incoming wave of high-density plasma, Houston sends an immediate alert to Orion. Even when the capsule is minutes away from entering the lunar blackout zone, the crew has enough time to halt their operations and move into the heavily shielded core of the capsule.
Back on Earth
When the Artemis II crew completes their lunar flyby and splashes down, the models guiding them will remain active. The same solar wind that threatens the astronauts also has the power to overwhelm high-voltage transformers, cause continent-wide electrical blackouts, and disable global GPS networks. To combat this, NASA made the computer code for the DAGGER model open source. Utility companies now receive that same 30-minute warning.
Leave a Reply