Auroral electrojet predictions with dynamic neural networks

Neural networks with internal feedback from the hidden nodes to theinput [Elman, 1990 are developed for prediction of the auroralelectrojet index AE from solar wind data. Unlike linear and nonlinearautoregressive moving-average (ARMA) models, such networks are free todevelop their own internal representation of the recurrent statevariables. Further, they do not incorporate an explicit memory for paststates; the memory is implicitly given by the feedback structure of thenetworks. It is shown that an Elman recurrent network can predict around70 of the observed AE variance using a single sample of solar winddensity, velocity, and magnetic field as input. A neural network withidentical solar wind input, but without a feedback mechanism, onlypredicts around 45 of the AE variance. It is also shown that fourrecurrent state variables are optimal: the use of more than four hiddennodes does not improve the predictions, but with less than that theprediction accuracy drops. This provides an indication that theglobal-scale auroral electrojet dynamics can be characterized by a smallnumber of degrees of freedom.