In a context of worldwide decline and given the critical ecological role of marine seagrasses to coastal ecosystem structure and functioning, regional conservation initiatives have emerged over the past thirty years to protect these important habitat-forming species. Yet, effective interventions need to account for site-specific processes and stressors. Thus, our ability to accurately predict seagrass dynamics is pivotal to support management interventions. To date, determinist process-based modelling has provided important insights on the drivers of seagrass dynamics. Here, we developed an original model framework that combines a coastal hydrodynamics ocean model with local data-driven models that rely on Boosted Regression Trees to predict seasonal dynamics of patch-level and plant-level seagrass features as a function of site-specific environmental conditions. Based only on a 12-month monitoring across nine sites, seagrass traits models successfully reproduce overall seasonal dynamics based mostly on inferred relationships with monthly light and temperature, and to a lesser extent, exposure to physical stressors (i.e., currents and waves). While models fail to finely capture spatial discrepancies across all sites (especially where seagrass demonstrates higher growth potential), spatially-explicit simulations highlight how seagrass-hydrodynamics feedback across the whole bay can dampen seagrass potential for growth due to exposure to shear stress. However, this original framework offers the potential to simulate long-term changes in the extent and status of seagrass meadows in Arcachon Bay, explicit resolving hydro-sediment dynamics effects on light appears as a priority to better capture the range of feedback processes between seagrass and coastal environmental conditions.