Earthquake occurs at the end of the seismic cycle. During a long time, the plate tectonics increase the stress loading along the plate boundaries, where mega-subduction earthquake eventually happen. Today, thanks to permanent GPS stations installed on subduction margins, we can measure the deformation of the crust associated with the subduction interface loading. This observation can be used to estimate in advance a crucial parameter of the earthquake in preparation: the stress drop. Knowing in advance the distribution of one of the two important parameters controlling the rupture enhances greatly the predictability of the future earthquake. By carefully evaluating the other parameters (the friction parameters), it is possible to compute future rupture scenarios, which can help to mitigate tsunami and strong-motion hazards. This new technique has been recently developed for the Nankai subduction zone, in Japan. However, a real earthquake did not validate the prediction technique. This is the first objective of this project: we need to apply the same technique to a monitored subduction zone where an earthquake occurred, to serve as a blind test. We will then compare the prevision made using only interseismic studies with the actual rupture. The discrepancies will be studied in order to improve the anticipation quality. This closely related to segmentation issues. What is responsible for the apparent segmentation of subduction zones? Can segmentation change over several seismic cycles? Since our goal is to be able to determine reliably what will be the maximum extent of the future rupture, we need to study very large earthquakes, which occurred in sufficiently instrumented regions before the earthquake. There are two equivalent places in the world where we can develop this project: Japan, with the 2011 Mw9.0 Tohoku earthquake, and Chile, with the 2010 Mw8.8 Maule earthquake. The method was initially developed for the Nankai trough, and we have started the study of Tohoku earthquake recently, but Chile seems a better target for this initial part of the project, since ENS team has been working on Chile for years now, and have a stronger expertise than for Japan needed to understand the limitations of the modelling. Of course, developments for Chile will have a strong impact in Japan, and are likely to be applied there too. The final task will be to elaborate scenarios for identified gaps close to rupture in Chile, and in Japan. Around the promising hopes of this approach, there are limitations that we want to investigate deeply. A second objective of this project is to model more precisely the interseismic deformation, and the corresponding stress accumulation on the slab. So far, simplified crustal models, and purely elastic behaviour are used to constrain the slip-deficit distribution. The elastic hypothesis is certainly valid for crustal behaviour at the time of the earthquake, but is it still a good hypothesis when looking at the long time interseismic loading? A related task is to investigate the role of spatial heterogeneities of friction at small scale on both earthquake preparation phase and the rupture. What is the meaning of the so-called plate-coupling rate? Are there different kinds of barriers? Can we locate barriers as well as asperities? There are several points we want to investigate which could improve our fundamental understanding of the mechanical processes leading to megathrust rupture. Finally, the project will contribute to one of the great challenge of modern seismology: rupture dynamic inversion. This technique is not well developed due to several limitations. We propose to improve the method. We will modify the dynamic inversion algorithm to take advantage of the interseismic information. Introducing a realistic a priori stress drop distribution will reduce the number of inverted parameters and increase the reliability of the solutions. Here again, both Tohoku and Maule earthquakes are good candidates.