Title: Efficient sampling strategies for non-Gaussian data assimilation

Author: Ahmed Attia (Virginia Polytechnic Institute and State University)
Vishwas Rao (Virginia Polytechnic Institute and State University)
Adrian Sandu (Virginia Polytechnic Institute and State University)

Data assimilation combines information from models, measurements, and priors to estimate the state of a dynamical system such as the atmosphere. Two strategies are generally followed to solve the data assimilation problem, namely variational and ensemble-based methods. Most of the data assimilation algorithms can handle linear and weakly nonlinear observation operators efficiently. Handling high nonlinearity of observation operators is usually a challenging task. Also, the underlying errors, e.g. background errors, and observation errors, are usually assumed to be Gaussian for the data assimilation algorithm to produce a useful analysis. We describe a general set of ensemble-based data assimilation methods, which obtains the analysis by sampling directly from the posterior distribution. The sampling strategy is based on a Hybrid Monte Carlo (HMC) approach that can handle non-Gaussian probability distributions. Two versions are presented: a HMC sampling filter for sequential data assimilation, and a HMC sampling smoother for four dimensional data assimilation. Numerical experiments are carried out with Lorenz-96 model, and with the shallow water model on a sphere. The numerical results explain the power of the sampling algorithms and suggest that they can outperform the traditional methods in nonlinear non-Gaussian settings.

GMAO Head: Steven Pawson Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: Feb 9 2015