Brief introduction to OpenDA native components

Current practice in data assimilation software

Data assimilation techniques are widely used in various modeling areas like meteorology, oceanography and chemistry. Most implementations of data assimilation methods however are custom implementations specially designed for a particular model. This is probably a consequence of the lack of generic data assimilation software packages and tools. An advantage of these custom implementations is that they are in general very computationally efficient. But the use of custom implementations has a number of significant disadvantages:

Costs - The development and implementation of these methods is very time consuming and therefore expensive.
Incompatibility - It is very hard to reuse these data assimilation methods and tools for other models than they were originally developed for.

How OpenDA improves the situation

The OpenDA project tries to enhance the reuse of data assimilation software by offering a modular framework for data assimilation, containing methods and tools that can be easily applied for general applications. OpenDA is set up in order to be as computationally efficient as possible, without losing its generic properties. The aim is that applications developed with OpenDA have a comparable computational performance as custom implementations.

OpenDA offers support for both users and developers of data assimilation methods. For users it allows models to be quickly connected to the OpenDA framework and hence to all the methods that are available in OpenDA. For developers, OpenDA offers efficient basic building blocks that save a lot of programming work and at the same time makes the new data assimilation software directly connectable to all OpenDA compliant models.

How OpenDA works

OpenDA provides a generic framework for data assimilation. It is aimed both at model programmers that want to use data assimilation methods and at developers of data assimilation methods.

OpenDA for model programmers

For model programmers, OpenDA provides a rich set of data assimilation methods. To use them, you have to make your model OpenDA compliant. Once your model can interact with OpenDA, all the OpenDA methods are at your disposal.

What is OpenDA compliant?

Making your model OpenDA compliant involves implementing a number of routines that are going to be called from the data assimilation methods. The set of routines that you must implement is called the stochastic model interface. There are other interfaces as well, each one defining a specific entity called a component. You will read more about components in the next section.

How OpenDA calls your implementation of the interface routines

OpenDA connects your implementations of these methods to their standard names. These standard names are used in the implementation of the data assimilation methods. There are provisions for working with black-box models(i.e. models for which you do not have the source code), but this will not be discussed in this document.

Providing your observations

Likewise, your observations must be provided in a OpenDA compliant way. For the observations there is also a set of routines, called the stochastic observer interface. Usually, you will not implement the interface but convert your observations to a format that can be handled by the standard OpenDA implementation of the stochastic observer component.

OpenDA for developers of data assimilation methods

For developers of data assimilation methods, OpenDA offers a platform for quickly building data assimilation methods that can be used from a wide range of models. OpenDA provides various sets of routines that can be used as building blocks. Such a set is called a component. You will read more about components in the section Components.

Routines to interact with models

First of all, OpenDA specifies a set of routines to interact with the model to which the data assimilation must be applied. Each OpenDA compliant model implements these routines (otherwise it will not be OpenDA compliant). The specification of this set of routines is called the stochastic model interface.

Routines to interact with observations

A second set of routines that is essential for data assimilation methods comprises routines to interact with the observations. These include routines to retrieve values and routines to retrieve meta-information.

Basic building blocks

Finally, there are various sets of basic building blocks (e.g. for handling vectors, matrices and time). These interact seamlessly with the other OpenDA components to let you construct data assimilation methods with a minimum of coding.