In data assimilation, one assumes the existence of a observation operator $\mathcal{H}$ that maps the model-state vector $\bf{x_b}$ to $ \bf{y_b}$ (the model-equivalent of the observations $\bf{y_o}$) according to a reference I'm using to develop a preliminary understanding of DA.

Can someone please elaborate on the precise meaning of:

model-equivalent of the observations $\bf{y_o}$

and the methods one can use to estimate the operator $\mathcal{H}$.


Think of $\mathbf{H}$ as a matrix operator which allows you to extract the part of the modelled state vector, which is observable. Just to give you a trivial example - assuming your state vector consists of 2D location and two velocity components and what you observe is just 2D location (e.g. obtained from GPS): $$\mathbf{y_o} = \left( \begin{matrix} x_o(t)\\ y_o(t)\\ \end{matrix} \right)$$ $$ \mathbf{x_b} = \left( \begin{matrix} x(t)\\ y(t)\\ v_x(t)\\ v_y(t) \end{matrix} \right) $$ $$ \mathbf{H}=\left( \begin{matrix} 1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\ \end{matrix} \right) $$ To check how $\mathbf{H}$ works, calculate on a sheet of paper the following products: $\mathbf{H x_b}$ and $\mathbf{H^T y_o}$. To construct a proper $\mathbf{H}$ operator you need to understand both what are you observing and what are you modelling.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.