Аннотация:This study is based on 10 years of ongoing GPS measurements of postseismic deformation following a pair of great earthquakes in the Kuril subduction zone. The study area is one of only a handful to capture a magnitude >8 signal and its postseismic signature with continuous GPS. The Kuril GPS Array was installed a few months prior to a 2006/2007 earthquake doublet and has been recording postseismic displacements in the near and far fields since that time. For a decade, the near field stations are moving trenchward, towards the seismic source at a speed of several tens of millimeters per year initially and an order of magnitude slower currently. Our modeling of viscoelastic relaxation explores realistic 3D subduction structures accounting for the dipping slab and for a low-viscosity mantle wedge above it (software RELAX of S. Barbot). We test linear (Maxwell) and nonlinear (power-law) rheologies of the asthenosphere, assuming that the viscoelastic relaxation is the dominant signal compared to afterslip after a year since the earthquakes. The data are best fit by the Maxwell asthenospheric viscosity 1 1018 Pa s for an interval 2007.52016.5. This viscosity is about ten times smaller than for two M9 events (Chile 1960 and Alaska 1964) from postseismic GPS displacements observed several decades later. This suggests a power-law rheology predicting the growth of apparent viscosity with time. From laboratory experiments with olivine, two alternative power-law mechanisms are possible: dislocation creep (stress power-law exponent n = 3.4-4.5) or diffusion creep (n = 0.9-1.5). Our numerical tests spanned the expected range of n, as well as a range of values of the initial apparent viscosity. The data are best fit by diffusion creep with n = 1.2 although the fit is not as good as for the Maxwell model.