We examined two techniques for eliminating components induced by variations of the geomagnetic field to accurately and quickly detect anomalous changes in the geoelectric field, which would enable us to identify crustal signals,. One method is based on a multiple regression model, where any geoelectric variation is presented as a linear combination of the past, present, and future values of geomagnetic variations. The induced geoelectric variation is estimated as a convolution of the geomagnetic variations and the parameters of the obtained regression model. The other method uses BAYTAP-G, by which the observed geoelectric data are separated into four components: 1) tidal, 2) electromagnetically induced, 3) trend, and 4) irregular components. Signals due to volcanic and/or seismic activity may be detected since the trend and the irregular component indicate variations of the self-potential of the crust and/or noise, provided that the amplitudes exceed that of the typical variation. The geomagnetic field at the Kakioka Magnetic Observatory was used as associated data for both methods. These methods were applied to geoelectric variations in the Numazu group, which are contaminated by artificial noise but show anomalous changes. The multiple regression method can apparently eliminate daily variations and telluric storms and can clarify anomalous changes that are not obvious in the original data. However, the gain and phase characteristics calculated from the estimated parameters of the model do not yield information on the underground resistivity structure since the estimated impulse responses do not correctly reflect relationships between the inducing geomagnetic field and the induced geoelectric field due to the presence of excessive artificial noise. BAYTAP-G separates daily variations due to artificial noise as the tidal component and the responses to geomagnetic variations. Anomalous changes are identified in the trend and the irregular component. We investigated the relationship of the tidal component of geoelectric variations estimated by BAYTAP-G with the tidal current and the tide. The correlation between the tidal component of geoelectric variations and tidal currents could not be clarified since simultaneous data of tidal currents with geoelectric variations were not available. Temporal variations of the amplitudes of geoelectric variations of KSM-MTO and tides at OAR for O1 and M2 constituents were not correlated despite their high coherencies.
We propose a time-domain method for estimating the apparent resistivity and phase. The method utilizes a multiple regression model where the order is determined by minimizing the AIC (Akaike Information Criterion). The method is superior to conventional frequency-domain methods based on spectral analysis, since the data length necessary for the latter methods is several times longer than that for the former in estimating the apparent resistivity at the same low frequency. Therefore, applying this method to shorter length data having long wavelengths like those in a magnetic storm, we can estimate apparent resistivities and phases at low frequencies with little effect of the finite wavelength of the inducing geomagnetic field. This method was applied to geoelectric field data observed by means of telegraphic facilities and geomagnetic field data at the Kakioka Magnetic Observatory, which is the standard observatory in Japan, located in the observation network of the geoelectric field. The geoelectric field data are very stable over a long time period because the electrodes are buried to a depth of more than 5 meters and have a contact resistance of less than 2 ohms. Furthermore, the geoelectric data had a high signal-to-noise ratio (signal is induced variations), because the lengths of observation lines ranging from 18.8 to 27.4 km were much longer than that of ordinary observations and the greater part of geoelectric variations were induced by geomagnetic variations. Apparent resistivity and phase in the low-frequency range from 3.3 × 10-5 to 10 × 1.0-3 Hz (corresponding to periods of 512 to 16 minutes) were estimated. Since the time span of the data used in the analysis was within the period of a large-scale magnetic storm, the estimates were unlikely to be affected by the finite wavelength of the inducing geomagnetic field.