Snowpack failure and snow avalanche release involve intrinsically multi-scale mechanisms, in which the instability of a whole mountain slope is intimately tied to rupture processes and reorganizations occurring at the scale of snow grains. When subjected to large strain rates, snow features a quasi-brittle mechanical behaviour characterized by strain-softening and volumetric collapse. Discrete Element (DEM) simulations based on microtomographic images of real snow samples have shown that this mechanical response is strongly sensitive to the specific microstructure of the material [1], which can vary in a large range. To tackle the challenge of modelling large-scale phenomena such as avalanches while capturing the complex mechanical behaviour of the material and the influence of the microstructure, a double-scale MPMxDEM approach is proposed. The MPM (Material Point Method) solver, which is used to compute the deformation and flow at large scale, is coupled to a homogenized numerical constitutive law stemming from the resolution of numerous individual DEM problems at each material point. Hence, each macroscopic lumping point embeds its own microstructure that can evolve independently. At the micro-scale, loose assemblies of spheres with a cohesive contact law are considered. The microscopic parameters are adjusted to recover the macroscopic stiffness and strength of the material. The two-way coupling strategy between the two numerical methods will be explained. Preliminary benchmark numerical simulations will then be presented and qualitatively compared to experimental and field data. The simulated configurations involve oedometric compression tests, Propagation Saw tests (PST), and the failure of a layered snowpack under its own weight.