Abstract

Adaptive Sampling, Data assimilation and Adaptive Modeling

Pierre F.J. Lermusiaux
Harvard University

Data assimilation allows the efficient acquisition
of multidisciplinary data via adaptive sampling
and efficient selection of model dynamics via adaptive modeling. Quantitative adaptive sampling uses data-driven dynamical forecasts and their uncertainties to predict in space and time
the observation system that is optimum for regional coverage, dynamical study and/or uncertainty reduction. Multidisciplinary adaptive sampling is especially challenging because of the multiple interdisciplinary correlations. Quantitative adaptive modeling uses model-data misfits and their uncertainties to evolve and select the model dynamics that are most suited to the rapidly evolving ocean dynamics. Both structural as well as parametric adaptation are possible. Conceptual issues and ongoing methodological developments, as well as computational and numerical considerations, will be outlined. Research with real ocean data is carried out with the Harvard Ocean Prediction System (HOPS) and Error Subspace Statistical Estimation (ESSE) data assimilation system. Real-time uncertainty forecasting, data assimilation, adaptive sampling, dynamical investigations, multi-model estimation and adaptive biogeochemical modeling will be illustrated for
the Autonomous Ocean Sampling Network-II (AOSN-II)
experiment in Monterey Bay.
Environmental-acoustical uncertainty estimation and transfers, and coupled data assimilation for physical-acoustical-seabed predictions and inversions will be illustrated in the Middle Atlantic Bight shelfbreak front region (PRIMER).