Flood inundation presents substantial societal hazards and also changes biogeochemistry for systems like the Amazon. It is often expensive to simulate high-resolution flood inundation and propagation in a long-term watershed-scale model. Due to the Courant-Friedrichs-Lewy (CFL) restriction, high resolution and large local flow velocity both demand prohibitively small time steps even for parallel codes. Here we develop a parallel surface-subsurface process-based model enhanced by multi-resolution meshes that are adaptively switched on or off. The high-resolution overland flow meshes are enabled only when the flood wave invades to floodplains. This model applies semi-implicit, semi-Lagrangian (SISL) scheme in solving dynamic wave equations, and with the assistant of the multi-mesh method, it also adaptively chooses the dynamic wave equation only in the area of deep inundation. Therefore, the model achieves a balance between accuracy and computational cost.