AbstractAn advanced data assimilation system, the local ensemble transform Kalman filter (LETKF), has been interfaced with a Regional Ocean Modeling System (ROMS) implementation on the Chesapeake Bay (ChesROMS) as a first step toward a reanalysis and improved forecast system for the Chesapeake Bay. The LETKF is among the most advanced data assimilation methods and is very effective for large, nonlinear dynamical systems with sparse data coverage. Errors in the Chesapeake Bay system are due more to errors in forcing than errors in initial conditions. To account for forcing errors, a forcing ensemble is used to drive the ensemble states for the year 2003. In the observing system simulation experiments (OSSEs) using the ChesROMS-LETKF system presented here, the filter converges quickly and greatly reduces the analysis and subsequent forecast errors in the temperature, salinity, and current fields in the presence of errors in wind forcing. Most of the improvement in temperature and currents comes from satellite sea surface temperature (SST), while in situ salinity profiles provide improvement to salinity. Corrections permeate through all vertical levels and some correction to stratification is seen in the analysis. In the upper Bay where the nature-run summer stratification is −0.2 salinity units per meter, stratification is improved from −0.01 per meter in the unassimilated model to −0.16 per meter in the assimilation. Improvements are seen in other parts of the Bay as well. The results from the OSSEs are promising for assimilating real data in the future.