Reconstructing Magnetotail Reconnection Events Using Data Mining is Feasible and Repeatable

Abstract Recently, Stephens et al. (2023), https://doi.org/10.1029/2022ja031066 utilized a data mining (DM) algorithm, applied to 26 years of magnetospheric magnetometer observations coupled with a flexible formulation of the magnetospheric magnetic field, to reconstruct the global configuration of the magnetotail when the Magnetospheric MultiScale (MMS) mission observed tail reconnection in situ. Of the 26 DM‐reconstructed MMS reconnection events, 16 had a isocontour within Earth radii of the observed reconnection location. Another eight had a minimum region, identified using nT isocontours, within . This consistency suggests that the structure of tail reconnection is correlated with the substorm/storm state of the magnetosphere, as reflected by geomagnetic indices and solar wind conditions. We verify these results using new validation methods and by comparing in‐sample (including event data) to out‐of‐sample (excluding event data) reconstructions. We first benchmark the architecture of the reconstructed magnetic field using 100 randomly generated magnetic fields containing tail X‐ and O‐lines, resolving 77 of them with three false positives. Next, we quantify the consistency of the reconstructions in resolving the reconnection location using a skill score relative to random chance. 88% of the in‐sample and 75% of the out‐of‐sample scores are positive, confirming that the reconstructions resolve the location of tail reconnection better than random chance. Last, a bootstrapping analysis, which refits the model architecture to 100 random resamples of data, shows standard deviations in of nT, indicating that the DM approach is not overly sensitive to the particular sampling of magnetometer records.