A new algorithm for classification of sea ice types on Sentinel-1 Synthetic Aperture Radar (SAR) data using a convolutional neural network (CNN) is presented. The CNN is trained on reference ice charts produced by human experts and compared with an existing machine learning algorithm based on texture features and random forest classifier. The CNN is trained on two datasets in 2018 and 2020 for retrieval of four classes: ice free, young ice, first-year ice and old ice. The accuracy of our classification is 90.5% for the 2018-dataset and 91.6% for the 2020-dataset. The uncertainty is a bit higher for young ice (85%/76% accuracy in 2018/2020) and first-year ice (86%/84% accuracy in 2018/2020). Our algorithm outperforms the existing random forest product for each ice type. It has also proved to be more efficient in computing time and less sensitive to the noise in SAR data. The code is publicly available.

Over the last decade, three major subduction earthquakes, Maule M_w 8.8 (2010), Illapel M_w 8.3 (2015) and Iquique M_w 8.1 (2014), occurred in Chile and generated significant post-seismic deformations. These large scale and long lasting deformations can be quantified with modern GNSS precise positioning and highlight viscoelastic processes in the asthenosphere. Here, we calculate the ratios of cumulative post-seismic displacements after 5 yr over the coseismic offsets. We find that at any distance from the trench, ratios are similar for the three earthquakes despite their different magnitudes which imply induced stresses that are more than one order of magnitude apart. This observation suggests that the post-seismic deformation is related to the same effective viscosity for the three earthquakes, indicating Newtonian rheology, rather than power-law rheology in the asthenosphere.

Over the last 3 decades, GPS measurements have been instrumental in quantifying tectonic plates current motion and deformation. Complex patterns of deformation along the plate boundaries revealed heterogeneous coupling on the plates interface and imaged seismic segments at different stages of their seismic cycle. Along the South-American trench in Chile, where large earthquakes occur frequently, continuous GPS observations (cGPS) captured both the long-term plate motion and the transient deformations associated with the seismic cycle. Over the years, a network of hundreds of cGPS stations has been deployed all across the South-American continent by many different institutions for many different sorts of purposes ranging from geographic reference to tsunami early warning. We report here on the processing of 20 years (2000–2020) worth of data over a selection of cGPS stations, devoted to the quantification and analysis of the deformation along the Chilean subduction zone between 18°S and 40°S. We use all available data near the trench in Chile and a less dense network inside the continent where the gradient of deformation is lesser. Our database, namedSOAM_GNSS_solENS, provides time series of precise daily station position, obtained from double difference (DD) processing and expressed in the International Terrestrial Reference Frame 2014 (ITRF14). These time series allow to quantify, with sub-millimetric precision, any kind of ongoing deformation process, either from tectonic origin such as interseismic deformation, co- and post-seismic displacements associated with earthquakes, transient deformation associated to seismic swarms and/or a-seismic slow-slip events, or of other origin such as hydrological loading (for ex., the Amazonian basin load) or any other type of loading affecting the surface of the Earth (tides, atmosphere, etc.). We also provide a database of coseismic displacements associated with close to 60 earthquakes of Mw larger than 6.5 that occurred over the last 20 years within the observation area. All time series are directly accessible through a deposit and we plan to make them available through a web interface that will allow any user to perform elementary operations like estimating offsets, detecting outliers, detrending, filtering and stacking. That database will evolve with time, aggregating more data. In the future, we also plan to complement that database with a rapid solution in quasi real time processed in Precise Point Positioning (PPP), and with hourly atmospheric delays associated with water vapor content of the lower layer of the atmosphere.

En induisant soudainement de fortes variations de contraintes en profondeur, les séismes de subduction s’avèrent être des expériences privilégiées pour étudier le comportement mécanique de l’intérieur de la Terre. Grâce au développement continu des réseaux d'observations GPS, il est possible de mesurer avec une précision millimétrique la déformation de la lithosphère au cours du cycle sismique. Durant la phase post-sismique du cycle, des déformations millimétriques sont ainsi mesurées sur de larges échelles temporelles (d'une année à plusieurs décennies) et spatiales (milliers de kilomètres). Ces déformations sont dues à la relaxation viscoélastiques des contraintes induites par les séismes dans l'asthénosphère.Ces dernières décennies, les modèles élastiques ont été largement utilisés pour reproduire les déformations à proximité de la fosse de subduction au cours du cycle sismique. Cependant, de tels modèles ne tiennent pas compte du caractère viscoélastique de certaines couches profondes de la Terre, et ne permettent donc pas de reproduire les déformations en champ lointain. Au cours de ce travail de thèse, j'ai donc cherché à analyser et démontrer l'importance des effets viscoélastiques sur la déformation modélisée de la plaque Amérique du Sud en réponse au cycle sismique le long de l’interface de subduction chilienne. J'ai également cherché à déterminer ce que les déformations mesurées par GPS nous apprennent sur la viscosité de l'asthénosphère.Pour ce faire, j'ai développé un nouveau maillage 3D sphérique en éléments-finis de la subduction chilienne couvrant la totalité du continent sud-américain. J'ai montré que la taille de la boîte englobante du maillage doit être largement supérieure à la taille de la zone étudiée. De plus, ce maillage intègre la subduction quasi horizontale en face de la région d’Illapel au Chili. Ces variations de pendage du slab doivent être prises en compte, car elles ont un impact sur les déformations mesurées par GPS. Je l'ai démontré en comparant les déformations verticales post-sismiques des séismes de Maule et d’Illapel. En comparant les déformations co- et post-sismiques générées par les trois grands séismes chiliens du 21ème siècle, j'ai montré que le fluage dans l'asthénosphère est linéaire (Newtonien). Cette linéarité du fluage permet de décomposer la déformation du continent sud-américain comme la somme de la déformation induite par chacun des segments de la subduction chilienne. J'ai également montré, que dans le cadre de la modélisation viscoélastique, la déformation actuelle est impactée non seulement par le séisme le plus récent, mais également par ceux qui l'ont précédé. Grâce à ces modèles, j'ai montré que la viscosité effective de l'asthénosphère varie au cours du cycle sismique et nécessite d’être modélisée par des rhéologies linéaires dites multi-Burgers, mettant en série un élément de Maxwell et plusieurs éléments de Kelvin-Voigt.

Axial melt lenses (AMLs) are key features of fast and magmatically-robust spreading ridges. These sill-shaped bodies typically sit atop a lower crustal mush zone, and supply magma that gets intruded in the brittle axial lithosphere as dikes, or emplaced at the seafloor as lava flows. The replenishment rate of these shallow reservoirs is thus a critical control on the modes of crustal accretion, the timing of mid-ocean ridge eruptions, and the thermo-chemical output of hydrothermal convection, but remains scarcely documented.Here we revisit estimates of magmatic inflation rates at the East Pacific Rise, 9º50’N based on measurements of vertical seafloor displacements carried out by Nooner et al. (2014). These measurements revealed seafloor uplift rates as fast as ~7 cm/yr above the AML, decaying over ~10 km in the cross-axis direction, between 2009 and 2011. We model this uplift profile as resulting from the inflation of a 1.5 km-deep, 1-km wide AML in a visco-elastic half-space that includes a viscous mush zone of uniform viscosity.Our models reveal a tradeoff between the assumed viscosity of the mush zone and the sill inflation rate that is necessary to explain the observed seafloor uplift. Specifically, if we assume a strong mush (viscosity > 1018 Pa.s), the replenishment rate must be ~200 m3/yr per meter along axis. On the other hand, a weaker mush (viscosity < 1016 Pa.s) significantly damps the surface expression of sill inflation, requiring a replenishment rate of ~470 m3/yr/m to match the observations. Further constraints on AML replenishment rates can be obtained by assuming the associated heat flux sustains on-axis hydrothermal venting near 9º50’N (~100 MW). We also find that rapid AML deflation during an eruption can induce a characteristic deformation transient lasting up to a few years, which is akin to the post-seismic phase of the seismic cycle. Depending on the assumed viscosity of the mush zone, this post-eruption signal has the potential to bias estimates of steady AML replenishment rates.Regardless of the assumed mush viscosity, our modeling yields replenishment rates comparable to the long-term crustal accretion rate (~600 m3/yr/m). This suggests that magmatic inflation is not an unusual event at a fast-spreading ridge like the East-Pacific Rise. By estimating the fraction of the ridge’s magma supply that transits through the AML, our results may also provide new constraints on the modes of accretion of the oceanic lower crust, i.e., help discriminate between the gabbro glacier and multiple-sills endmember models.