A multilayer feed‐forward neural network model has been developed to forecast >2 MeV electron flux at geosynchronous orbit. The model uses as input 10 consecutive days of historical electron flux values and 7 consecutive days of daily summed values of the planetary Kp index with two neurons in a single hidden layer. Development of the model is discussed in which the size of the training set interval and the retraining period are investigated. Problems associated with neuron saturation which limit the ability of the network to generalize are shown to be circumvented through a daily retraining regimen. The model performance is evaluated for the period 1998–2008 and compared with the results produced by the REFM model. The neural network model is demonstrated to perform quite well relative to the REFM model for this time period, producing mean prediction efficiencies for 6 month test intervals of 0.71, 0.49, and 0.31 for 1 day, 2 day, and 3 day forecasts, respectively.

We examine the relationship of energetic equatorial electron flux enhancements occurring near L -4.2 and 6.6 associated with 26 well-defined high-speed solar wind streams (HSSWS) detected by Wind between December 1994 and Septelnber 1996. Events were selected fbr having high-energy (>2 MeV) geosynchronous electron daily average fluxes surpassing 103 cm -2 s -1 sr -1 for at least a day as measured by GOES 7 or GOES 9. Los Alamos differential-energy electron data from SOPA (0.2 -2.0 MeV) at L -6.6 and the GPS BDD-II dosimeters (0.2 -3.2 MeV) at oe -4.2 illustrate that flux dropouts are typically observed in all energy channels at both equatorial altitudes within the first day of each event. While SOPA consistently records postdropout fittx enhancements, GPS dosimeters detect equatorial postdropout enhancements in 1.6-3.2 MeV electron fluxes in only 15 of 26 events and all are either concurrent (1 event) with or follow (14 events) the geosynchronous increases of electrons with similar values of the first adiabatic invariant,/t -• 2.1 x 103 MeV G -1. In addition, 10 of 15 GPS growth periods produced electron enhancements above predropout levels. For all 26 events the phase space density for electrons of similar 3t is consistently greater at geosynchronous altitude than at GPS equatorial altitude. The critical factor leading to GPS L = 4.2 electron flux enhancements is elevated geomagnetic activity levels of Kp --, 3.0 -3.5 and above for extended periods. A combination of enhanced solar wind ram pressure, electric field (with B-south), and velocity also appears to be necessary. If outward phase space density gradients are combined with the large electric fields generally accompanying elevated Kp, then sufficient conditions may exist to promote the inward radial diffusive transport of equatorial electrons that ultimately lead to electron flux enhancements at GPS altitudes. Comparison of observed and theoretically estimated electron growth rates is consistent with this picture of inward radial transport for these equatorially mirroring particles with 3t ~ 2.1 x 103 MeV G -1 at L -4.2.

A tilt-dependent magnetic field model of the Earth's magnetosphere with variable magnetopause standoff distance is presented. Flexible analytic representations for the ring and cross-tail currents, each composed of elements derived from the Tsyganenko and Usmanov (1982) model, are combined with the fully shielded vacuum dipole configurations of Voigt (1981). The ring current, consisting of axially symmetric eastward and westward currents fixed about the dipole axis, resembles that inferred from magnetic field observations yet permits easy control of inner magnetospheric inflation. The cross-tail current contains a series of linked current sheet segments which allow for the tilt-dependent flexing of the current sheet in the x-z plane and arbitrary variations in current sheet position and intensity along the length of the magnetotail. Although the current sheet does not warp in the y-z plane, changes in the shape and position of the neutral sheet with dipole tilt are consistent with both MHD equilibrium theory and observations. In addition, there is good agreement with observed AB profiles and the average equatorial contours of magnetic field magnitude. While the dipole field is rigorously shielded within the defined magnetopause, the ring and cross-tail currents are not similarly confined, consequently, the moders region of validity is limited to the inner magnetosphere. The model depends on four independent external parameters, namely, (1) the dipole tilt angle, (2) the magnetopause standoff distance, (3) the midnight equatorward boundary of the diffuse aurora, and (4) the geomagnetic index Dst. In addition, we present a simple but limited method of simulating several substorm related magnetic field changes associated with the disruption of the near-Earth cross-tail current sheet and collapse of the midnight magnetotail field region. These include the classic dipolarization of the near-Earth field and the reduction of the far-tail equatorial field accompanying current sheet thinning. This feature further facilitates the generation of magnetic field configuration time sequences useful in plasma convection simulations of real magnetospheric events. IntroductionA wide variety of currents are required to support the magnetospheric magnetic field structure in the presence of flowing solar wind plasma and the interplanetary magnetic field. In addition to the magnetic field generated in the Earth's interior, the major contributors to this structure include magnetopause currents, the ring and cross-tail currents, and a variety of field-aligned currents. The intensities of these currents fluctuate constantly as they feed into each other to form the global circuit. For this reason, a model of magnetic fields and currents must represent a diverse array of magnetospheric configurations as well as the "average" configurations. It must rely on physical magnetospheric and solar wind parameters and reproduce appropriate configuration changes. These changes include the distortion of magnetic field and its mapping characteristics ...

No abstract

1] Extended periods of relativistic electron intensity at geosynchronous orbit can create severe deepcharging hazards for satellites. Over the last 20 years a number of models have been developed to predict electron flux levels using solar wind and geomagnetic indices as inputs. We analyze the results of several of these including the Relativistic Electron Forecast Model based on the linear prediction filter technique, a neural network algorithm, and the physics-based diffusion method. Analyses using the methods of simple persistence and recurrence (based on the 27 day solar rotation) are also included as performance baselines. Comparisons are made to the GOES > 2 MeV electron flux to determine which model or method gives the best +1, +2, and +3 day forecasts of average daily flux during the interval 1996-2008. Model inputs include combinations of SK p , the daily average solar wind speed, and daily average > 2 MeV electron fluxes for one day or multiple days prior to the forecast days of interest. Prediction efficiencies are calculated for 6 month intervals. After evaluating all the models, there was no clear winner; each model did well at different phases of the solar cycle. All models perform their best during the inclining phase of solar minimum but not as well during solar maximum and the declining phase of solar minimum. While persistence is respectable for +1 day prediction, models clearly give superior +2 and +3 day predictions and should be used to obtain those forecasts.