Data assimilation with stochastic reduced models quantifying model error
Abstract
Data assimilation combines data with dynamical models to make prediction, using ensemble of solutions to represent the uncertainty. Due to limited computational resources, coarse-grid models are often used, but model errors from the sub-grid scales often prevent accurate predictions. Therefore, it is important to account for the model errors and derive reduced models capturing the key statistical-dynamical properties. We propose a stochastic parametrization method which accounts for the model error by nonlinear autoregression moving average (NARMA) type models. We show that the reduced model faithfully describes the distribution of the underlying process of a linear hypoelliptic SDE. We demonstrate by examples that the resulting NARMA type stochastic reduced models can capture the key statistical-dynamical properties and therefore can improve the performance of ensemble prediction in data assimilation. The examples include the Lorenz 96 system and the Kuramoto-Sivashinsky equation of spatiotemporally chaotic dynamics.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMNG33B0950L
- Keywords:
-
- 3315 Data assimilation;
- ATMOSPHERIC PROCESSESDE: 1910 Data assimilation;
- integration and fusion;
- INFORMATICSDE: 3275 Uncertainty quantification;
- MATHEMATICAL GEOPHYSICSDE: 4468 Probability distributions;
- heavy and fat-tailed;
- NONLINEAR GEOPHYSICS