| description abstract | Continuous monitoring by satellites of the ocean spatiotemporal variability allows empirical forecasting of satellite-observed data. The forecast of satellite data is achieved in three major phases: decomposition of the spatiotemporal variability of the satellite data, noise reduction of the encoded space, and time variability and the prediction, by a nonlinear forecasting technique, of the time variability. Different techniques can be employed in each processing phase. Specifically, decomposition can be carried out using spatial or temporal variance EOF decomposition. With spatial variance EOFs, performance of the empirical forecasting is associated with the predictability of the dynamics of spatial structures with strong spatial gradients (fronts, eddies, etc.?.?.?.). Conversely, if temporal variance decomposition is employed, the success of the empirical prediction system will be related to features with strong temporal variability. Both approaches have been employed to empirically forecast satellite-observed data in different ocean areas. This article attempts to determine which is the superior method (in terms of predictability). Two satellite-based ocean forecasting systems using temporal and spatial variance decomposition, respectively, were developed to forecast the monthly mean of the sea surface temperature (SST) of the Adriatic Sea. Results indicate that, although the empirical dynamical models were different, both approaches provide similar forecasts. Slight differences in the forecast skill of both systems are found in spring and summer showing with better performance from the covariance SOFT system. | |
