An Efficient Data Assimilation Algorithm with a Gaussian Covariance Structure
Abstract
Details are provided of our attempts to develop computationally efficient and widely acceptable data assimilation algorithms. Our initial data assimilation work was done with a scheme which was developed by Derber and Rosati (1989) for their global data assimilation application. The scheme is quite efficient: but when it was combined with a primitive equation, numerical model of the Gulf of Mexico. the nowcasting step took too much computer time for it to be useful in real-time applications. So, after an initial implementation and test of the scheme. our obvious thrust was to enhance and make it more efficient. Most of the computation resources while using the Derber-Rosati scheme are required in approximating a matrix product Sigma mg, where Sigma m is the covariance matrix of the model output error having a Gaussian spatial covariance structure that stays fixed, and g is a varying vector. They approximate this product by repeated applications of a Laplacian operator. A new algorithm is developed that, while maintaining the accuracy, approximates Sigma mg far more efficiently than the Laplacian algorithm. It also admits a wider class of covariance structures with equal ease and efficiency. The new algorithm approximates Sigma mg by applying a product of two operators that are polynomials in simple averaging operators. The algorithm is efficient and general with the only restriction that Sigma m be based on a covariance function that is a product of factors, which are even functions in a single dimension. Data assimilation, Computationally efficient, Covariance function, Gaussian, Laplacian, Separable functions, Operator polynomials
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1993
- Accession Number
- ADA273152
Entities
People
- John C. Derber
- Kent Goodrich
- Mark Limber
- Ranjit M. Passi
Organizations
- University of Southern Mississippi