Earth is surrounded by a magnetic bubble called the magnetosphere, which protects us from harmful solar particles. At times, this protective shield temporarily opens up through a process called magnetic reconnection, a fundamental process in space plasmas that still has much scope for discovery and deeper understanding. These events are important because they can disrupt satellites, communications, and even power grids.

However, identifying reconnection events in space data is difficult because they often look very different from one another, can be hidden by turbulence, and there is a lack of properly labeled data available for training detection tools. This lack of labeled data is a major bottleneck for building reliable automated detection systems, as machine learning models require large and diverse training sets to achieve high accuracy.

We develop a way to create realistic “synthetic” examples of these events using physics-based rules. By generating these examples, we can train computers to better recognize a wide range of reconnection events in large space datasets.

This approach can help advance the study of these fundamental events more efficiently and improve space weather forecasting in the future.