A nonlinear filtering method is introduced for the study of the solar
wind - magnetosphere coupling and related to earlier linear techniques
. The filters are derived from the magnetospheric state, a representat
ion of the magnetospheric conditions in terms of a few global variable
s, here the auroral electrojet indices. The filters also couple to the
input, a representation of the solar wind variables, here the rectifi
ed electric field. Filter-based iterative prediction of the indices ha
s been obtained for up to 20 hours. The prediction is stable with resp
ect to perturbations in the initial magnetospheric state; these decrea
se exponentially at the rate of 30 min(-1). The performance of the met
hod is examined for a wide range of parameters and is superior to that
of other linear and nonlinear techniques. In the magnetospheric state
representation the coupling is modeled as a small number of nonlinear
equations under a time-dependent input.