Model error moment estimation via data assimilation

TL;DR

This paper introduces a novel method to estimate the mean and covariance of model errors over a single timestep using data assimilation, providing bounds based on state estimation errors, which aids in experimental design.

Contribution

It derives the first bounds on model error statistics estimation errors in terms of state estimation errors, linking observational information to model error quantification.

Findings

Provides bounds on errors in estimating model error mean and covariance.

Shows how state estimation accuracy influences model error estimation.

Facilitates experimental design by indicating observational information requirements.

Abstract

Using a dynamical model to make predictions about a system has many sources of error. These can include errors in how the model was initialised but also errors in the dynamics of the model itself. For many applications in data assimilation, probabilistic forecasting, or model improvement, these model errors need to be known over the timestep of the model, not over a time-averaged period. Using a forecast from a state that combines observational information as well as prior information we can gain an approximation to the statistics of the model errors on the timescale of the model that is required. Here we give bounds on the errors in the estimation of the mean and covariance of the errors in the model equations in terms of the errors made in the state estimation. This is the first time that such a result has been derived. The result shows to what extent the state estimation must constrain the analysis in order to obtain a specified error on the mean or covariance of the model errors. This is particularly useful for experimental design as it indicates the necessary information content required in observations of the dynamical system.