Empirical Reconstruction of Pre‐1995 Extreme Storms Using ML‐Derived Solar Wind Inputs

Abstract The storm‐time geomagnetic field and electric currents are reconstructed for extreme storms before 1995: the July 1982 superstorm and the March 1989 Hydro‐Québec grid collapse event. The reconstructions are based on an improved magnetic field data mining method utilizing recently published machine learning‐derived solar wind data. The data mining reconstructions are rescaled using statistics of the nearest neighbor bins to eliminate the bias toward weaker storms. A concurrent reconstruction method provides the combined description of storms and substorms: storm and substorm features are first reconstructed independently for the inner and tail magnetosphere, respectively, and then the data fitting is reiterated using synthetic data generated using the first round of reconstructions. The data fitting procedure is further tuned to better resolve the location of the field‐aligned currents. Testing the updated methods for the November 2003 and 1982 superstorms significantly improves the validation results for in situ observations. The effect of rescaling doubles the peak ring current density (from 81 to 168 nA/m2 for the November 2003 storm) while the tuned fitting procedure shifts the Region‐2 field‐aligned currents equatorward to magnetic latitudes as low as 50°. Rescaling also intensifies the equatorial currents such that X‐line arcs and even an X‐loop are formed within geosynchronous orbit, where reconnection may approach a relativistic regime. Such a change in the field topology limits the peak plasma pressure obtained from the quasi‐static force balance equation.