A systematic analysis of the plausibility of different classes of models which could account for low-frequency precursory electrotelluric signals observable at large distances from a large earthquake is carried out. First, we consider ''linear'' models for which either a stationary electrical field is generated by a current dipole in the source volume, or a static deformation field results from a precursory strain in the source volume, linearly inducing a streaming potential effect near the electrode. In the former case, only the conjunction of very favorable circumstances may lead to an electrical field larger than 10 mV/km at more than 100 km, with the constraint that no anomaly exceeds a few volts per kilometer in the source area: It indeed requires a powerful electrical dipole (current 104 A; length 1 km) which might be produced by a streaming potential effect associated with the connection of nonhydrostatic fluid reservoirs (pore pressure difference up to 100 MPa), a two-dimensional channeling of the electrical currents from the source to the electrode area, which would result from high conductivity horizontal layers, and finally a local amplification factor of typically 100 near the electrodes, which might be produced by a narrowing of the conductive channel. For an elastic long distance interaction, an electric field of 10 mV/km at a distance 10 times the source length (100 km away from a M=6 earthquake) requires a precursory deformation larger than the coseismic deformation, if no strong local amplification is considered.

Alternatively, ''nonlinear'' models are proposed and discussed, assuming that the crust is stuffed with small scale local instabilities (fractures and nonhydrostatic fluid reservoirs), among which a number are close to the instability threshold. These instabilities may follow their own evolutionary process, inducing sporadic fluid flow and hence streaming potential effect in the vicinity of the electrodes, without any relation to a remote tectonic activity. However, these instabilities may be sensitive to very small changes in the stress field, and hence be triggered by the elastic precursory deformation of a remote seismic source. This requires that the precursory deformation near the instability is at least larger than the tidal stain (10-7); this necessary condition is met at a distance of up to 10 times the source dimension when the precursory strain source is only 10% of the coseismic deformation, provided that the strain source is double the size of the seismic fault itself. This model could account for the absence of self-potential anomalies related to the earthquake itself, as a large proportion of the available instabilities could have been ''swept away'' by the precursory strain. Furthermore, with such a model, a precursory strain in the source area could induce a detectable change in the background seismic activity over a vast area, which may give a new interpretation of the times of increased probability observations. For short distances, the triggering of fluid instabilities by strain perturbation and the associated streaming potential effect may well be a promising model for explaining low frequency electric and magnetic precursors. ¿ American Geophysical Union 1992