Sensitivity of sub-regional ocean forecasts to different initialization techniques

In the present work we analyze the quality of the ocean forecasts produced in the Central Mediterranean area as function of different initialization techniques. The model producing the forecasts is a Princeton Ocean Model implementation for the Central Mediterranean, namely the Sicily Channel sub-Regional Model (SCRM). Model is forced at surface with 1 -hourly atmospheric forecasts and at open boundaries receives information from a coarser basin scale model. In the present study forecasts are initialized in three different ways , namely slave mode (S), partially-active (PA) and active (A) mode. While in S mode the initial condition is created by interpolation of scalar and vector fields from the coarse model (also providing boundaries), in partially-active and active modes the model has more time to develop its own dynamics. In PA mode the model performs a two-weeks hindcast to create initial conditions, during which it may partially develop its own dynamics. In A mode forecasts are initialized from a continuous analysis run assimilating Sea Level Anomalies through a 3D-variational software. Forecasts quality for the three modes is verified against remotely sensed SST. Both optimally interpolated SST as well "raw" synoptic SST were used. For each forecast "cycle" basic statistics and some non-dimensional skill have been calculated. Results provides interesting information on the best initialization procedures to be adopted in operational forecasting when using one-way nested models.