Hindcasting and Forecasting of the POLYMODE Data Set with the Harvard Open–Ocean Model

Abstract

Abstract A regional quasi-geostrophic model has been used to hindcast and forecast the POLYMODE data set. After briefly discussing hindcast methodology, the hindcast fields are compared with the analyzed data set Periods of significant difference of hindcast from analysis are identified and investigated. We find that these differences may be largely attributed to inaccuracies in the analysed fields. The inaccuracies are due to a lack of data. When the data set fails to adequately describe the ocean, the hindcast may be more accurate than the analyzed fields. Model studies also demonstrate that hindcast quality improves after being degraded by a period of poor boundary conditions, topographic forcing is relatively important in improving the accuracy of the hindcasts, and idealized numerical resolution studies are applicable to the assimilation of oceanic data sets. Methods for forecasting are examined and intercompared for several periods during the POLYMODE experiment. Forecast accuracy is found to be highest when statistical techniques are used to forecast the boundary conditions and the interior evolves as determined by dynamics. Away from boundary condition induced errors, the dynamical model is able to maintain a high level of correlation between the forecast and analyzed fields for 20 days. Also, the accuracy may be affected by the position of the data relative to the forecast domain. The implications for sampling strategies an discussed. Thew results are important to ocean scientists on several fronts. In studying mesoscale processes, a continuous time series of fields may be important for analysis of the kinematics and dynamics. When conducting a field measurement program, knowledge of evolving mesoscale fields may aid in the positioning of sensors. These topics are briefly discussed and future plans described.