Advanced Assimilation Strategies in Modern Observational Networks for Real-Time, High Resolution Applications

Rapid advances in computer technology and estimation/simulation algorithms over the last few years have made possible the melding of multiplatform observations with models in real time using advanced and versatile assimilation methodologies. The error subspace statistical estimation (ESSE) scheme is used to assimilate the various data gathered by the observational network at the Long-Term Ecosystem Observatory (LEO-15) into a coupled atmosphere-ocean (RAMS/ROMS) forecast system along the Southern New Jersey coast. The ESSE is a nonlinear, multivariate minimum variance approach based on the dominant optimal reduction of the dimension of the time-variant, state error covariance matrices. A Monte Carlo perturbation ensemble method is used to determine the evolution and adaptive learning of the dominant variability and error covariances. Up to 300 perturbations, per assimilation cycle, are needed in this application (with a state vector of about 2.75 million points) to achieve convergence of the error subspace forecast. Massive parallel computers are used to assimilate new data, predominantly surface velocity data from CODAR, every six hours. The error subspace approach is used to determine the adaptive sampling and better use of the observational assets.