The apparent magnetic noise, obtained from the coherency function for two parallel magnetic sensors, generally overstimates sensor noise because the sensors do not measure the same signal. The different signals result from the nonparallel alignment of the sensors and from the additional magnetic signal induced in each sensor by its motion in the Earth's magnetic field. A magnetometer array experiment was completed in Grass Valley, Nevada, to determine the minimum magnetic signal that could be detected in the presence of background natural field variations and motion of the sensor. Superconducting quantum interference device (SQUID) magnetometers with interval biaxial tiltmeters were used to record the magnetic fields and the motion of the sensors.

A least squares fitting program enabled the field at one site in the array to be predicted from a remote site and to produce a residual field with a standard deviation of 8 pT over a 1-hour period with a low-pass filter at 0.3 Hz. Consistent field-to-residual ratios of 40--60 dB were achieved, with some ratios exceeding 70 dB. The least sauares fit uses only a linear combination of the magnetic and tilt fields at a remote site to predict the observed magnetic field. This procedure allows for correction of calibration and orientation errors as well as the removal of the apparent fields originating from sensor movement. Misalignment and motion of the sensor are shown to be the major sources of magnetic field noise. The orientation error is typically of the same magnitude as the noise induced by sensor motion. In order to achieve ratios better than 20--40 dB one must include both the orthogonal fields and the tilmeter outputs. Inclusion of a frequency-dependent transfer function should increase the prediction ability of the least squares model, as evidenced by the improvement to 70 dB obtained with simple band limiting of the original data. These techniques should be applicable to type of artificial source survey where natural field fluctuations are the noise-limiting factor. The ability to describe the observed signals should allow a dramatic increase in one's ability of detect an artificially generated signal, allowing signal-to-noise improvements of 40--60 dB without increasing transmitter power or the averaging time. ¿ American Geophysical Union 1988