Weather predictability at the mesoscale, and its sensitivity to the initial conditions and observing systems, is being evaluated over the New Zealand region using a mesoscale resolving assimilation and forecasting system. The New Zealand Limited Area Model (NZLAM) is a regional scale implementation of the mesoscale configuration of the Met Office Unified Model (UM). The horizontal grid spacing for the NZLAM is12 km and it is configured on a 324 x 324 x 38 level grid centred over the Tasman Sea. Data assimilation is carried out using 3DVAR, and the full system (NZLAM-VAR) is cycled 3 hourly. Boundary conditions are provided by running the UM in global weather prediction mode. These systems (UM, NZLAM and NZLAM-VAR) are providing opportunities to evaluate the relative and absolute skill of global (UM), non-assimilating high-resolution interpolator (NZLAM), and assimilating (NZLAM-VAR) prediction systems – and hence to evaluate the impact of mesoscale data assimilation over a domain with significant orography, but dominated by a marine environment. The paper will comment on the results of the data impact studies to date, with an emphasis on the utility of satellite observations (e.g. sea surface temperature, and surface wind speed, as well as HIRS and AMSU sounder radiances) and the formulation of the NZLAM-VAR forecast error covariance matrix.