Bridging in Situ Satellite Measurements and Simulations of Magnetic Reconnection Using Recurrent Neural Networks

Magnetic reconnection is inherently structured, with distinct spatial regions such as inflows, outflows, and separatrices playing key roles in energy conversion and particle transport. While in situ spacecraft measurements provide detailed local information, determining where a spacecraft lies within the global reconnection geometry remains a major challenge. Proxy-based methods are often ambiguous, while full reconstructions require strong assumptions and are difficult to apply systematically across events. Here, we present a method that bridges these approaches by using machine learning to infer global structural context from local measurements. We first apply k-means clustering to a 2.5-D particle-in-cell simulation to identify six characteristic symmetric reconnection regions. A recurrent neural network (RNN) is then trained on spacecraft-like trajectories through the simulation to classify time series data into these regions. When applied to Magnetospheric Multiscale (MMS) observations of magnetotail reconnection, this method successfully identifies regional transitions, including inflow, outflow, and separatrix crossings, in agreement with previous reconstructions where available. The approach provides a practical, scalable, and automated framework for determining spatial context in reconnection events without requiring full geometric reconstruction, enabling large-scale and efficient statistical studies of reconnection dynamics across multiple events.