Sequential data assimilation methods for atmospheric general circulation models

Abstract

The data assimilation (DA) process has gained some spotlight in recent years as computers have become more powerful, and models more complex. Even so, most natural phenomena have many correlations among variables that are very challenging to capture. In this proposal, we discuss the impact of an intermediate step in the leaping strategy used as a numerical integrator for Atmospheric General Circulation Models during the assimilation process, and its explicit update, particularly, for the Simplified Parameterizations, privitivE-Equation DYnamics model, nicknamed as SPEEDY. Using literature validated formulations of the Ensemble Kalman Filters the Local Ensemble Kalman Filter (LEnKF), Local Ensemble Transform Kalman Filter (LETKF), and the Ensemble Kalman Filter based on a Modified Cholesky Decomposition (EnKF-MC) experimental test are performed using the leaping step in the update process, and using only the forecast step, and letting the model propagate the updates. For the EnKF-MC formulation, we propose a formulation onto the observations space. As well, we present an intuitive Python package to perform sequential data assimilation on atmospheric general circulation models. We denote our package by Applied Math and Computer Science Lab - Data Assimilation AMLCS-DA. This package contains the efficient implementations of the previously mentioned formulations. The results reveal that our proposed framework can properly estimate model variables within reasonable accuracies in terms of Root-Mean-Square-Error when we update only the forecast state, even when using sparse operational observators (25%, 11%, 6%, 4%).