This paper reports on a magnetotelluric (MT) survey across the central Mariana subduction system, providing a comprehensive electrical resistivity image of the upper mantle to address issues of mantle dynamics in the mantle wedge and beneath the slow back‐arc spreading ridge. After calculation of MT response functions and their correction for topographic distortion, two‐dimensional electrical resistivity structures were generated using an inversion algorithm with a smoothness constraint and with additional restrictions imposed by the subducting slab. The resultant isotropic electrical resistivity structure contains several key features. There is an uppermost resistive layer with a thickness of up to 150 km beneath the Pacific Ocean Basin, 80–100 km beneath the Mariana Trough, and 60 km beneath the Parece Vela Basin along with a conductive mantle beneath the resistive layer. A resistive region down to 60 km depth and a conductive region at greater depth are inferred beneath the volcanic arc in the mantle wedge. There is no evidence for a conductive feature beneath the back‐arc spreading center. Sensitivity tests were applied to these features through inversion of synthetic data. The uppermost resistive layer is the cool, dry residual from the plate accretion process. Its thickness beneath the Pacific Ocean Basin is controlled mainly by temperature, whereas the roughly constant thickness beneath the Mariana Trough and beneath the Parece Vela Basin regardless of seafloor age is controlled by composition. The conductive mantle beneath the uppermost resistive layer requires hydration of olivine and/or melting of the mantle. The resistive region beneath the volcanic arc down to 60 km suggests that fluids such as melt or free water are not well connected or are highly three‐dimensional and of limited size. In contrast, the conductive region beneath the volcanic arc below 60 km depth reflects melting and hydration driven by water release from the subducting slab. The resistive region beneath the back‐arc spreading center can be explained by dry mantle with typical temperatures, suggesting that any melt present is either poorly connected or distributed discontinuously along the strike of the ridge. Evidence for electrical anisotropy in the central Mariana upper mantle is weak.

S U M M A R YGeomagnetic induction responses such as geomagnetic depth sounding (GDS), magnetotelluric (MT), and horizontal geomagnetic transfer function (HTF) at long periods are used to estimate the electrical conductivity in the deep mantle. The responses in the period range that are shorter than 10 5 s (about 1 day) are in many cases considered to be local or regional induction problems in which the source field is approximated by plane waves and therefore the sphericity of Earth is not taken into account. In the period range between 10 4 and 10 5 s, the most dominant signature of the magnetic field variation is the solar quiet daily (Sq) variation and its higher harmonics. Therefore, when we obtain the responses due to the quasi-white background spectrum composed of plane waves, we regard the Sq field variations as noises in estimating the responses, and line spectra of the variations are removed from the observed time-series before the responses are calculated. However, with this approach, the calculated responses tend to possess a discontinuity at a period of about 10 4 s, and the response functions show common features at longer periods irrespective of the location of the observation site. It is particularly well known that the imaginary part of the induction vector tends to have a significant westward component for periods ranging between 10 4 and 10 5 s. Such features cannot be easily explained by the effect of the electrical conductivity structure alone. Examination of the phase of the HTF implies that the responses in the same period range are affected by the signature of sources of finite wavelength moving westward. Thus, it is suggested that the response functions in this period range were under the effect of the Sq field variations, even though the line spectra of them were removed and the responses were estimated at periods separate from the harmonic periods of Sq field variation. We examined how the influences of the Sq field appear on the responses numerically by a forward modelling. Results show most of the characteristic features in observed response functions can be ascribed to Sq source effects.