Water circulation between atmosphere and geosphere generates important feedbacks. Thus, a coupled model has been developed by combining the Multi-Scale Simulator for the Geoenvironment for Atmosphere (MSSG-A), for Ocean (MSSG-O) and the GEneral purpose Terrestrial fluid-FLOW Simulator (GETFLOWS) for surface-groundwater. The information exchange between the simulators is controlled dynamically by the OASIS coupler. The coupled model can handle short time (10min to daily), seasonal (several months to yearly) and potentially decadal (not implemented yet) water circulation. The model is used to investigate the influence of temperature, rainfall and evaporation on land surface run-off and ground water as well as soil moisture and latent heat on climate. An important application is to model the impact of future climate change and support sustainable development. The current paper focus on model description and mass-balance assessment. Some preliminary experiments were performed using a model with a 500m horizontal resolution, exchanging data every 10 minutes on 147 computing cores. The model covers the whole Kanto basin in Japan and is assessed over the period 21/07/2006 to 08/09/2006 that includes one extreme rainfall event. First, the water mass balance of the coupled model was estimated by comparing the aquifer water storage (groundwater, soil moisture and river storage) computed by the hydrological model and the water flux of the watershed such as rainfall, evaporation, river discharge and coastal exchange at the land-ocean interface. The coupled model achieved good mass conservation. In a second step, the present coupled model was validated with observation data. Comparison with rainfall station data showed that the model underestimated the rainfall at high elevation locations but overestimated the rainfall at low elevation locations. Despite these biases, the overall rainfall was well reproduced by the model. Comparison with gridded satellite derived TRMM rainfall data showed that the cumulative precipitation volume modelled in the domain agrees with the satellite observation with an error of 15% or less.